TWI884731B - Ai recognition and recovery device for acupuncture needles - Google Patents

Abstract

The present invention relates to an AI recognition and recovery device for acupuncture needles, which includes: a frame; a needle placement platform for placing a plurality of acupuncture needles; a driving unit provided on the frame; a photography unit provided with a lens pointing toward the needle placement platform for photographing the acupuncture needles and thereby obtaining an acupuncture needle image; a computing unit for receiving the acupuncture needle image and generating a binary image through calculation; an AI module for recognizing the number of acupuncture needles from the binary image and outputting a recognized value, wherein the AI module is created by deep learning based on a training dataset established through employing a YOLO object detection model and a recursive gated convolution network to perform two rounds of high-level and low-level feature fusion; and an accommodation space provided on the frame. When the recognized value meets the total number of the acupuncture needles used during acupuncture, the driving unit collects the acupuncture needles into the accommodation space.

Description

AI Identification and Recycling Device for Acupuncture Needles

The present invention relates to an AI recognition system that can improve the accuracy and speed of recovering acupuncture needles.

At present, in general TCM clinics, doctors usually use acupuncture to treat patients. The number of acupuncture needles used each time is usually between dozens, and the number of outpatients at the same time period is quite large. Therefore, after each patient's acupuncture treatment, doctors occasionally miss a needle when removing it, which usually causes the patient to be injured due to the failure to remove the needle. In severe cases, it may even endanger the patient's life. Therefore, it is required that doctors record the number of needles and the area of the body when performing acupuncture. After removing the needles, they need to check whether the number of acupuncture needles is consistent with the record. However, since the acupuncture needles need to be counted one by one and added up, it is quite labor-intensive and time-consuming, and thus delays the consultation time of the next patient. In addition, although there is a method of using metal electromagnetic induction to sense whether there are still acupuncture needles left in the patient, this method can avoid missing acupuncture needles that have not been removed, but it also requires manpower and time to calculate and add up the number of acupuncture needles.

Therefore, there is a patent case of invention No. I680779 "Needle Recovery Box with Counting Device" announced on January 1, 2020 in the Republic of China. It discloses: for use with multiple needles, it includes a container and a counting module detachably connected to the container. The container includes a bottom plate, a surrounding wall erected around the bottom plate, and a storage space defined by the bottom plate and the surrounding wall. The counting module includes a delivery port connected to the storage space, a needle sensing device arranged corresponding to the delivery port, a counting display arranged facing away from the storage space, and a controller electrically connecting the needle sensing device and the counting display. The needle sensing device transmits a detection signal to the controller each time the needle enters the accommodating space, and the controller controls the counting display to display the corresponding number according to the detection signal.

When discarding needles, if the acupuncture needles overlap, the number of needles may be missed or misjudged, so the use of these prior patents is not ideal.

There is also a new patent case No. M635081 "Medical Automatic Recycling Device" announced on December 1, 2012. It discloses: It is an automated device for providing medical waste for classification and recycling, which includes a box body, including an upper space and a lower space, the upper space has an opening, and the upper space and the lower space are interconnected; a plurality of storage bins, which are mainly set in the lower space of the box body; a waste identification device, which is set in the box body, and includes a tilting plate, an image identifier and a controller, the tilting plate has a vertical rod, and the vertical rod is set near the lower space. The tilting plate is arranged at the end of the vertical pole, so that the height of the tilting device is slightly higher than the storage buckets and the vertical pole is used as the axis to perform 360-degree pivoting motion. The image recognition device is arranged in the upper space and is arranged toward the tilting plate. The controller is electrically connected to the image recognition device and the tilting device respectively, so that the image recognition of the medical waste on the tilting plate can be performed, and the tilting plate can be rotated to the corresponding storage bucket through the controller to perform the tilting operation and classification.

The patent application cannot be used to identify very small objects such as acupuncture needles, and therefore cannot be used to identify acupuncture needles.

Therefore, in view of the above-mentioned shortcomings of the current acupuncture needle recovery device, the present invention provides an AI recognition and recovery device for acupuncture needles, comprising: a frame; a needle placement platform, which is arranged on the frame and is used to place a plurality of acupuncture needles; a driving unit, which is arranged on the frame; a camera unit, which has a lens pointing to the needle placement platform, and the camera unit takes pictures of the acupuncture needles to obtain an acupuncture needle image; a calculation unit, which is connected to the camera unit, receives the acupuncture needle image, and obtains a binary image through calculation; an AI module, The signal is connected to the computing unit, the AI module uses the YOLO object detection model, utilizes the recursive gated convolutional network to perform secondary high-low order fusion, and is established by deep learning with a training data set; the AI module recognizes the number of the acupuncture needles from the binary image and outputs an identification value accordingly; a storage space is set in the frame; when the identification value meets the total number of needles, the driving unit gathers the acupuncture needles into the storage space.

The above further includes an electronic operation unit, which is signal-connected to the computing unit. The electronic operation unit executes an acupuncture needle management program. The acupuncture needle management program includes an input interface. The input interface is provided with at least one portrait acupuncture diagram. The portrait acupuncture diagram has a plurality of acupuncture areas. Each of the acupuncture areas is provided with a partitioned numerical area for inputting a number of acupuncture needles. The numbers of acupuncture needles in all the acupuncture areas are aggregated by the electronic operation unit into the total number of acupuncture needles and displayed in a total display area of the input interface.

The above further includes an electronic operation unit, which is signal-connected to the computing unit. The electronic operation unit executes an acupuncture needle management program. The acupuncture needle management program includes a timing module. The electronic operation unit is used to input an acupuncture time. The timing module executes a time record. When the time record matches the acupuncture time, the electronic operation unit outputs a prompt message.

The binary image contains a plurality of linear set pixels. The operation unit defines the binary image as a rectangular coordinate plane, takes the pixels of the binary image as coordinate units, and further performs a slope verification on the linear set pixels to calculate a pixel slope in each linear set pixel. When the pixel slopes are all within an error range, the operation unit determines that the acupuncture needle is a single one.

The needle placement platform can rotate relative to the frame, and the driving unit drives the needle placement platform to rotate, so that the acupuncture needles on the needle placement platform fall into the accommodating space by gravity.

The frame is provided with an identification space, the accommodating space is located below the identification space, the frame is provided with a bracket protruding toward the identification space, and the photographic unit is fixed to the bracket.

The needle placement platform is configured as a flat plate to be pivotally connected to the frame to separate the identification space and the accommodating space.

The storage space is provided with a door panel that can be opened or closed movably. A receiving bucket can be placed in the storage space. The receiving bucket is located below the identification space.

The needle placement platform is pivotally connected to the frame via a pivot shaft, and the driving unit is linked to the pivot shaft via a variable speed transmission assembly to drive the needle placement platform to rotate toward the accommodating space at a tilting angle.

The driving unit is a stepper motor, and the speed transmission set is a gear transmission set or a belt transmission set.

According to the above technical features, it has the following advantages:

1. After the acupuncturist has removed all the acupuncture needles from the patient, they can be placed on the needle placement platform and identification can be started at the same time. If the number matches, the acupuncture needles on the needle placement platform will be controlled to use gravity to fall into the receiving bucket in the storage space and be collected. In this way, the number of acupuncture needles can be quickly counted and cleared to improve the accuracy and speed of recovering the acupuncture needles.

2. At the same time, the calculation unit will feed back the above recognition results to the electronic operation unit to inform the acupuncturist whether the number of needles removed is consistent with the total number of needles. If the number is found to be inconsistent, the subsequent needle removal or needle search can be carried out immediately to reduce the occurrence of missed acupuncture needles.

3. By operating the electronic operation unit to execute the acupuncture needle management program, a portrait needle diagram can be displayed to facilitate the acupuncturist to perform acupuncture according to the corresponding part of the patient's body. The acupuncture site and the number of needles can be selected through the portrait needle diagram in the corresponding acupuncture area, and the corresponding number of needles can be input, so that the total number of needles can be quickly summarized to facilitate subsequent identification and comparison.

4. The timing module is operated to execute timing, and when the waiting time for the needle to be left in is reached, the electronic operation unit will emit a warning light or a warning sound to remind the acupuncturist to prepare to remove the needle, so as to avoid the needle being left in for too long and causing harm to the patient.

5. For the improvement of AI modules, the YOLO object detection model (HPS-YOLOv7) is mainly used. It is an algorithm specially suitable for high-precision small object detection. It can be used to identify small objects such as acupuncture needles to achieve accurate identification.

6. An improved efficient layer aggregation network (CLAN) is proposed for feature extraction, which not only retains the network depth but also reduces the number of parameters, making the recognition model more lightweight and helping to improve the recognition speed.

7. The recursive gated convolution (Cg ⁿ Conv) was improved and used to replace the original Neck's CBS structure, making the model more flexible and avoiding deviations. It avoids the asymmetry caused by high-order convolutions, not only retains smaller objective information, but also does not bring additional parameters, making it more helpful in identifying small objects.

8. A shallow feature fusion network (SFN) is proposed to combine deep convolution with shallow semantic information, making up for the lost information of small objects in the deep convolution network. The 20*20 detection head is changed to 160*160, thereby improving the detection accuracy of small objects and increasing the training speed.

9. Increasing the amount of data in the training dataset can better simulate good recognition rates in different situations. It is necessary to continuously collect data on acupuncture needles, and also consider various possible situations from simple single needles to multiple needles and then to overlapping needles, so as to facilitate the deep learning of the training model.

10. Use the slope to assist in identification and help determine whether the head and tail of the covered acupuncture needle belong to a single needle or different needles. Based on various data surfaces such as the length, width, and placement direction of the acupuncture needle, these parameters are put into the set slope for the AI module to distinguish and use the point-slope method to calculate the slope between the pixels of the acupuncture needle. This can distinguish between two overlapping and different acupuncture needles, thereby greatly increasing the overall recognition accuracy.

11. Image processing uses an adaptive binarization method, which is set according to the average value of the area of a specified size to reduce the influence of light and shadow factors, making the data received by the model cleaner, which is conducive to the accuracy of subsequent recognition. Then, various needle types are mixed and overlapped, and this pre-processing action is repeated. The needles in each situation can be identified cleanly in sequence, thereby greatly improving the recognition accuracy.

Please refer to the first, second and third figures, the embodiment of the present invention includes: a frame 1, a receiving space 2, a needle placement platform 3, a driving unit 4, a camera unit 5, a computing unit 6, an AI module 7 and an electronic operation unit 8, wherein:

The frame 1 is provided with an identification space 11. The frame 1 is provided with a bracket 12 protruding toward the identification space 11.

The storage space 2 is disposed in the frame 1. The storage space 2 is located below the identification space 11. The storage space 2 is provided with a door panel 21 that can be opened or closed movably. A receiving bucket 22 can be placed in the storage space 2. The receiving bucket 22 is located below the identification space 11.

The needle placement platform 3 is disposed on the frame 1 for placing a plurality of acupuncture needles A (as shown in the fourth figure). The needle placement platform 3 is configured as a flat plate and is pivoted to the frame 1 via a pivot 31, and can rotate relative to the frame 1 to separate the identification space 11 and the accommodation space 2.

The driving unit 4 is disposed on the frame 1. The driving unit 4 is a stepping motor, and the driving unit 4 can further drive the pivot 31 of the needle placement platform 3 through a speed change transmission set 41, and the speed change transmission set 41 is a gear transmission set or a belt transmission set. After the acupuncture needles A are identified, the driving unit 4 can drive the pivot 31 through the speed change transmission set 41 to drive the needle placement platform 3 to rotate toward the receiving space 2 at an inclined angle, so that the plurality of acupuncture needles A on the needle placement platform 3 can fall into the receiving bucket 22 in the receiving space 2 by gravity and be collected.

The camera unit 5 is fixed on the support 12 and has a lens 51 pointing to the needle placement platform 3. The lens 51 of the camera unit 5 photographs the acupuncture needles A to obtain an acupuncture needle image (as shown in the fifth figure).

The operation unit 6 is signal-connected to the camera unit 5 and the drive unit 4, respectively. The operation unit 6 receives the acupuncture needle image and obtains a binary image after calculation (as shown in Figure 6). The binary image refers to an image with only two pixel values, usually black and white. It is obtained by binarizing the grayscale image. In a binary image, each pixel can only be one of the two values, usually 0 represents one state (such as black) and 1 represents the other state (such as white). Binary images are often used in image processing and computer vision applications because images containing only two pixel values are easier to process and analyze. The acupuncture needle A in the present invention embodiment is black, and the background color is white. Such a representation helps to simplify the image and emphasize specific information to facilitate subsequent analysis and processing during identification. The binary image contains a plurality of linear set pixels. The operation unit 6 defines the binary image as a rectangular coordinate plane, and uses the pixels of the binary image as coordinate units. A slope verification is further performed on the linear set pixels to calculate a pixel slope in each linear set pixel. When the pixel slopes are within an error range, the operation unit 6 determines that the acupuncture needle A is a single one.

The AI module 7 is signal-connected to the operation unit 6. The AI module 7 uses the YOLO object detection model, which is a neural network-like algorithm, and uses a recursive gated convolutional network to perform secondary high- and low-level fusion, and is established by deep learning with a training data set; the AI module identifies the number of acupuncture needles A from the binary image and outputs an identification value accordingly. When the aforementioned identification value meets the total number of needles, the driving unit 4 gathers the acupuncture needles A into the accommodation space 2.

The electronic operation unit 8 is signal-connected to the operation unit 6. The electronic operation unit 8 is a smart phone or a tablet computer. The electronic operation unit 8 executes an acupuncture needle management program (APP), and the acupuncture needle management program includes an input interface 81, and the input interface 81 is a touch screen. The input interface 81 is provided with at least one portrait acupuncture diagram 82, and the portrait acupuncture diagram 82 has a plurality of acupuncture areas 83, and each of the acupuncture areas is provided with a partitioned numerical area 84 for inputting a number of acupuncture needles. The numbers of acupuncture needles in all the aforementioned acupuncture areas 83 are summarized by the electronic operation unit 8 as the total number of acupuncture needles, and the total number of acupuncture needles is displayed in a total display area 85 of the input interface 81. Furthermore, the acupuncture needle management program includes a timing module 86. The electronic operating unit 8 is used to input an acupuncture time. The timing module 86 executes a time record. When the time record matches the acupuncture time, the electronic operating unit 8 outputs a prompt message. The prompt message can be a warning light or a warning sound.

When in use, as shown in the first, second and third figures, after an acupuncturist (doctor) performs acupuncture on a patient, the acupuncturist also executes the acupuncture needle management program on the electronic operation unit 8, and displays the acupuncture image 82 of a front body portrait and a back body portrait through the input interface 81. According to the acupuncture on the corresponding part of the patient, the acupuncture parts and the number of acupuncture on the patient are selected one by one through the acupuncture image 82 in the corresponding acupuncture areas 83, and the corresponding number of needles is input. At the same time, the acupuncture time is input through the electronic operation unit 8, for example, the acupuncture time of 20 to 50 minutes after acupuncture. Then the electronic operation unit 8 activates the timing module 86 to execute the time record. At the same time, the acupuncture needle management program will summarize the needle numbers inputted in the acupuncture portrait 82 and the acupuncture areas 83 to form the total number of acupunctures, and display it in the total display area 85. At the same time, the total number of acupunctures will be transmitted to the calculation unit 6 for the calculation unit 6 to identify and compare according to the total number of acupunctures. When the time record matches the acupuncture time, the electronic operation unit 8 will output a prompt message, which can be a warning light or a warning sound, to remind the acupuncturist to prepare to remove the needles, so as to avoid leaving the needles for too long and causing harm to the patient.

As shown in the first and fourth figures, after the acupuncturist removes the acupuncture needles A from the patient, they can be placed one by one on the needle placement platform 3. When all the acupuncture needles A on the patient are removed and placed on the needle placement platform 3, the recognition can be controlled to start. First, the camera unit 5 will photograph the acupuncture needles A on the needle placement platform 3 to obtain an acupuncture needle image [as shown in the fifth figure]. The calculation unit 6 receives the acupuncture needle image and obtains the binary image through calculation [as shown in the sixth figure]. The binary image uses the binary method, so that the data received by the model is cleaner and the acupuncture needle A can be more prominent, which is beneficial to the accuracy of subsequent recognition. At the same time, it is necessary to mix and overlap various needle types in advance, repeat this pre-processing action, and sequentially identify the needles in each situation, thereby greatly improving the recognition accuracy. As shown in the fifth figure, it is the original image of the acupuncture needle image taken, and as shown in the sixth figure, it is the image after adaptive binarization. Due to the use of the adaptive binarization method, the required grayscale and threshold are set according to the average value of the area of the specified size, reducing the influence of light and shadow factors.

The AI module 7 uses the YOLO object detection model, uses a recursive gated convolutional network to perform secondary high- and low-level fusion, and is established by deep learning with a training data set. Since the embodiment of the present invention adopts the YOLO object detection model version 7 (HPS-YOLOv7), an algorithm specifically used for high-precision small object detection, it is very suitable for identifying small objects such as acupuncture needles A. An improved efficient layer aggregation network (CLAN) is proposed for feature extraction. The content of the improved efficient layer aggregation network lies in its network structure. As shown in Figure 7, it is a general residual network, and as shown in Figure 8, it is a bottleneck structure. Both use residual structures, but the difference lies in the number of channels. The original efficient layer aggregation network does increase the number of network layers, which helps improve the training accuracy and generalization ability of the model. However, multiple nonlinear layers increase the network depth, resulting in a huge amount of computation, which may lead to a decrease in network performance. Taking the seventh figure as an example, the general residual structure directly uses two 3*3 convolution kernels, and the bottleneck is about 1.17 million, while the bottleneck structure will only appear after the dimensionality is reduced by a 1*1 convolution kernel, and then after a 3*3 convolution kernel, and finally a 1*1 convolution kernel is used to increase the dimension, the total number of parameters is about 70,000, a difference of 16.7 times. It can be seen that the bottleneck structure not only retains the network depth, but also reduces the number of parameters, making the model more lightweight. Therefore, the embodiment of the present invention proposes the above-mentioned improved efficient layer aggregation network, and replaces the residual network with a bottleneck structure, which helps to shorten the time of recognition operation.

The embodiment of the present invention improves the recursive gated convolution (Cg ⁿ Conv) and is used to replace the original Neck's CBS structure. The difference between the original recursive gated convolution and the original recursive gated convolution in mathematical theory is that the original formula is:

P _{k +1} =ƒ k (q k )⊙ ɡ k (pk)/α, k =0, 1,…, n-1;

On this basis, a secondary high- and low-level fusion is performed, which becomes the following formula:

P _{k +1} ={[fk(qk)*gk(pk)]⊙[fn(qn)*pn]}/a, k =0, 1, ..., n-1;

We repeated this process several times. For the sake of scalability, we expanded the simple two-stage interaction to more stages to make the model more flexible and avoid bias. The recursive gated convolution (C-Recursively gated Convolution) avoids the asymmetry brought by high-order convolution. It not only retains smaller objective information, but also does not bring additional parameters, making it more helpful in identifying small objects.

The embodiment of the present invention proposes a shallow feature fusion network (SFN), which combines deep convolution with shallow semantic information to make up for the lost information of small objects in the deep convolution network. The 20*20 detection head is changed to 160*160, thereby improving the detection accuracy of small objects. Because the original YOLO object detection model version 5 (YOLOv5) uses a personal area network (PAN) to resample high-level feature maps and fuse previous feature maps to further improve detection accuracy, but the fusion of deep and shallow features will still lead to the loss of some semantic information. The shallow feature fusion network used here is to avoid over-sampling, which leads to the loss of small objects that are not obvious in themselves. Because acupuncture needle A is too thin and long, and is prone to dense problems, excessive deep sampling training will easily lead to the loss of the original features. Therefore, the reason for using this network structure is to retain the information of small and medium objects in the shallow convolution layer, improve the accuracy of small objects, and increase the training speed.

Since building a model requires increasing the amount of data in the training dataset, in order to better simulate and achieve good recognition rates in different scenarios, it is necessary to continuously collect data on acupuncture needle A. The data will also range from simple single needles to multiple needles, and then to overlapping needles, taking into account various possible situations to facilitate training the flexibility of the model.

The binary image contains a plurality of linear set pixels. The operation unit 6 defines the binary image as a rectangular coordinate plane, takes the pixels of the binary image as coordinate units, and further performs a slope verification on the linear set pixels to calculate a pixel slope in each linear set pixel. When the pixel slopes are all within an error range, the operation unit determines that the acupuncture needle A is single.

The above-mentioned part using the slope is used to assist in determining whether the head and tail of the covered acupuncture needle A belong to a single needle or different needles. Since the acupuncture needle A itself is a thin long cylinder, it will only have a partially overlapping characteristic, so the slope of each acupuncture needle A is different. According to the pixels of the image after image processing, the slope between the pixels in the needle is calculated using the point slope method, and two overlapping different needles can be distinguished. According to various data surfaces such as the length, width and placement direction of the acupuncture needle A, these parameters are added to the set slope, allowing the AI module 7 to distinguish, so that it can more effectively avoid the state where the needle head and needle tail are considered to be the same needle but are regarded as different needles by the model, thereby greatly increasing the accuracy of overall recognition.

The AI module 7 identifies the number of acupuncture needles A from the binary image and outputs the identification value to the calculation unit 6 accordingly. Then the calculation unit 6 compares the identification value with the number of needles inserted in different parts of the patient input by the acupuncturist and the total number of needles inserted after being summarized by the acupuncture needle management program. As shown in Figure 9, when the aforementioned identification value matches the total number of needles inserted, the calculation unit 6 will control the drive unit 4 to start, and drive the needle placement platform 3 to rotate the tilt angle toward the receiving space 2 through the speed change transmission group 41 to make the acupuncture needles A on the needle placement platform 3 fall into the receiving bucket 22 in the receiving space 2 by gravity and be collected. At the same time, the calculation unit 6 will transmit the above identification results to the electronic operation unit 8 to inform the acupuncturist whether the number of needles removed is consistent with the total number of needles. If the number is found to be inconsistent, the subsequent needle removal or needle search can be carried out immediately to reduce the occurrence of missing acupuncture needles A. In this way, through the AI identification and recovery device of the present invention, the number of acupuncture needles A can be quickly counted and the acupuncture needles A can be cleared and collected to improve the accuracy and speed of recovering the acupuncture needles A.

The present invention has been subjected to actual testing, and tests 1 to 6 have been conducted (the recognition image screenshots are shown in FIG. 10 to FIG. 15 respectively), and the test results are shown in Table 1 below.

Table 1: Test 1 Test 2 Test 3 Test 4 Test 5 Test 6 Total number of needles 44 44 44 44 44 44 Identify value 44 44 43 43 42 41 Recognition rate 100% 100% 97.7% 97.7% 95.4% 93.1%

From the results of the above-mentioned actual tests 1 to 6, it is known that the recognition rate of the present invention can reach at least 93.1%, and the average recognition rate is as high as 97.3%, which is sufficient to prove that the present invention can indeed achieve the effect of accurate and rapid recognition.

Combined with the description of the above embodiments, the operation, use and effects of the present invention can be fully understood. However, the above embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of the implementation of the present invention. In other words, simple equivalent changes and modifications made according to the scope of the patent application of the present invention and the content of the invention description are all within the scope of the present invention.

1: Frame 11: Identification space 12: Bracket 2: Accommodation space 21: Door panel 22: Receiver 3: Needle placement platform 31: Pivot 4: Drive unit 41: Variable speed transmission group 5: Camera unit 51: Lens 6: Calculation unit 7: AI module 8: Electronic operation unit 81: Input interface 82: Portrait acupuncture diagram 83: Acupuncture area 84: Partition value area 85: Total display area 86: Timing module A: Acupuncture needles

[Figure 1] is a three-dimensional appearance diagram of an embodiment of the present invention.

[Figure 2] is a combined cross-sectional view of an embodiment of the present invention.

[Figure 3] is a schematic diagram of the electronic operation unit and its input interface according to an embodiment of the present invention.

[Figure 4] is a schematic diagram of an acupuncture needle placed on a needle placement platform according to an embodiment of the present invention.

[FIG. 5] is a schematic diagram of an acupuncture needle image obtained by photographing an acupuncture needle by the camera unit of the embodiment of the present invention.

[Figure 6] is a schematic diagram of the image of the acupuncture needle after adaptive binarization according to the embodiment of the present invention.

[FIG. 7] is a schematic diagram of a general residual network of an efficient layer aggregation network according to an embodiment of the present invention.

[Figure 8] is a schematic diagram of the bottleneck structure of the high-efficiency layered polymerization network according to an embodiment of the present invention.

[Figure 9] is a schematic diagram of opening the needle placement platform to allow acupuncture needles to fall into the receiving bucket for collection according to an embodiment of the present invention.

[Figure 10] is a screenshot of the identification of Test 1 of the actual test of the present invention.

[Figure 11] is a screenshot of the identification of the second test of the present invention after actual testing.

[Figure 12] is a screenshot of the identification of the present invention in Test 3 after actual testing.

[Figure 13] is a screenshot of the identification of Test 4 of the actual test of the present invention.

[Figure 14] is a screenshot of the identification of Test 5 of the actual test of the present invention.

[Figure 15] is a schematic diagram showing the identification of Test 6 of the present invention after actual testing.

1:Frame

11: Identify space

12: Bracket

2: Accommodation space

21: Door panel

22: Receive bucket

3: Needle placement platform

31: Pivot

4: Drive unit

41: Speed transmission group

5: Photography unit

51: Lens

6: Arithmetic unit

7: AI module

8: Electronic operation unit

81: Input interface

82: Acupuncture portrait

Claims (10)

An AI identification and recovery device for acupuncture needles comprises: A frame; A needle placement platform, arranged on the frame, for placing a plurality of acupuncture needles; A driving unit, arranged on the frame; A camera unit, having a lens pointing to the needle placement platform, the camera unit photographs the acupuncture needles to obtain an acupuncture needle image; An operation unit, connected to the camera unit, receives the acupuncture needle image, and obtains a binary image through operation: An AI module, signal-connected to the computing unit, the AI module uses the YOLO object detection model, utilizes the recursive gated convolutional network to perform secondary high-low order fusion, and is established by deep learning with a training data set; the AI module identifies the number of the acupuncture needles from the binary image, and outputs an identification value accordingly; A storage space, set in the frame; When the identification value matches the total number of needles, the driving unit gathers the acupuncture needles into the storage space. The AI identification and recovery device for acupuncture needles as claimed in claim 1 further comprises an electronic operating unit, which is signal-connected to the computing unit, and which executes an acupuncture needle management program. The acupuncture needle management program comprises an input interface, and the input interface is provided with at least one portrait acupuncture needle diagram, and the portrait acupuncture needle diagram has a plurality of acupuncture needle areas, and each of the acupuncture needle areas is provided with a partitioned numerical area for inputting a number of acupuncture needles, and the numbers of acupuncture needles of all the acupuncture needle areas are aggregated by the electronic operating unit into the total number of acupuncture needles, and displayed in a total display area of the input interface. The AI identification and recovery device for acupuncture needles as claimed in claim 1 further comprises an electronic operating unit, which is signal-connected to the computing unit. The electronic operating unit executes an acupuncture needle management program. The acupuncture needle management program includes a timing module. The electronic operating unit is used to input an acupuncture time. The timing module executes a time record. When the time record matches the acupuncture time, the electronic operating unit outputs a prompt message. As in claim 1, the AI identification and recovery device for acupuncture needles, wherein the binary image contains a plurality of linear set pixels, the operation unit defines the binary image as a rectangular coordinate plane, takes the pixels of the binary image as coordinate units, further performs a slope verification on the linear set pixels, calculates a pixel slope in each of the linear set pixels, and when the pixel slopes are within an error range, the operation unit determines that the acupuncture needle is a single one. As in claim 1, the AI identification and recovery device for acupuncture needles, wherein the needle placement platform is rotatable relative to the frame, and the driving unit drives the needle placement platform to rotate, so that the acupuncture needles on the needle placement platform fall into the accommodating space by gravity. As in claim 1, the AI identification and recovery device for acupuncture needles, wherein the frame is provided with an identification space, the accommodating space is located below the identification space, the frame is provided with a bracket protruding toward the identification space, and the photographic unit is fixed to the bracket. As in claim 6, the AI identification and recovery device for acupuncture needles, wherein the needle placement platform is configured as a flat plate to be pivotally connected to the frame to separate the identification space and the accommodating space. As in claim 6, the AI identification and recovery device for acupuncture needles, wherein the storage space is provided with a door panel that can be movably opened or closed, and a receiving bucket can be placed in the storage space, and the receiving bucket is located below the identification space. As in claim 1, the AI identification and recovery device for acupuncture needles, wherein the needle placement platform is pivotally connected to the frame via a pivot, and the driving unit is connected to the pivot via a variable speed transmission assembly to drive the needle placement platform to rotate at an inclination angle toward the accommodating space. As in claim 9, the AI identification and recovery device for acupuncture needles, wherein the driving unit is a stepper motor, and the speed transmission set is a gear transmission set or a belt transmission set.