CN111476199A - Identification method of power transmission and transformation project cemetery based on high-definition aerial survey images - Google Patents

Abstract

The invention discloses a method for identifying a cemetery in a power transmission and transformation project based on high-definition aerial survey images. The method includes obtaining remote sensing image data; preprocessing the remote sensing image data to obtain training data; The network model is trained to obtain the cemetery identification model; the cemetery identification model is used to identify the remote sensing image data to be identified to obtain the identification result; the identification result is optimized to obtain the final cemetery identification result. The high-definition aerial survey image-based burial ground identification method for power transmission and transformation projects provided by the present invention realizes accurate, reliable and efficient identification of burial grounds through an innovative method of constructing a burial ground identification model; therefore, the method of the present invention has high reliability and high efficiency and high accuracy.

Description

Identification method of power transmission and transformation project cemetery based on high-definition aerial survey images

technical field

The invention belongs to the field of image processing, and in particular relates to a method for identifying a power transmission and transformation project graveyard based on high-definition aerial survey images.

Background technique

With the development of economy and technology and the improvement of people's living standards, electric energy has become an indispensable secondary energy in people's production and life, bringing endless convenience to people's production and life. With the development of society, the construction of power system and the selection of location and line of power transmission and transformation projects are particularly important.

In the traditional concept of our country, the concept of "grave" is very sensitive. If the site selection and line selection of the power transmission and transformation project (such as relocation of the cemetery, construction affecting the cemetery, the cemetery and its surrounding construction, high-voltage lines crossing the cemetery) involves the cemetery and surrounding construction, etc., it is very likely that residents and power system employees will be affected. conflict between. And even if the construction and wiring are successful, the equipment maintenance and repair in the later stage is also very difficult. Therefore, when selecting power lines, avoiding graveyards as much as possible is the first choice.

At present, in the process of site selection and line selection of power transmission and transformation engineering, the consistent practice is to set several schemes first, and then use manual calibration and identification to mark the cemetery area and information on the map, and then according to the calibration and marking methods Information is adjusted until a final solution is selected.

However, the manual calibration and labeling method is not only time-consuming and labor-intensive, and the accuracy is not high, but also there is the possibility of missed judgment and misjudgment, and the reliability is high and not high.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a power transmission and transformation engineering graveyard identification method based on high-definition aerial survey images with high reliability, high efficiency and high accuracy.

The method for identifying a power transmission and transformation project cemetery based on high-definition aerial survey images provided by the present invention includes the following steps:

S1. Obtain remote sensing image data;

S2. Preprocess the remote sensing image data obtained in step S1 to obtain training data;

S3. Build a neural network model for cemetery identification;

S4. using the training data obtained in step S2 to train the cemetery identification neural network model constructed in step S3, thereby obtaining a cemetery identification model;

S5. Use the cemetery identification model obtained in step S4 to identify the remote sensing image data to be identified, thereby obtaining an identification result;

S6. Optimizing the identification result obtained in step S5, so as to obtain the final identification result of the cemetery.

The step S2 preprocesses the remote sensing image data obtained in the step S1 to obtain training data, specifically, performing data enhancement on the remote sensing image data obtained in the step S1 and expanding the sample data to obtain the training data.

The building of the neural network model for cemetery identification in step S3 is specifically building a deep learning framework TensorFlow, so as to build a neural network model for identifying cemetery objects.

The neural network model is the Deeplabv3+ model.

The optimization processing in step S6 specifically includes noise point removal processing, smoothing processing and breakpoint connection processing.

The method for identifying a power transmission and transformation project cemetery based on high-definition aerial survey images further includes the following steps:

Model compression technology is used to compress the established recognition model, so as to improve the efficiency of the model.

The method for identifying cemeteries in power transmission and transformation projects based on high-definition aerial survey images provided by the present invention realizes accurate, reliable and efficient identification of cemeteries through an innovative method of constructing a cemetery identification model; therefore, the method of the present invention has high reliability and high efficiency. and high accuracy.

Description of drawings

FIG. 1 is a schematic flow chart of the method of the present invention.

FIG. 2 is a schematic diagram of the overall architecture of the Deeplabv3+ model in the method of the present invention.

FIG. 3 is a schematic diagram of hole convolution in the Deeplabv3+ model in the method of the present invention.

FIG. 4 is a schematic diagram of part of the Decoder in the Deeplabv3+ model in the method of the present invention.

FIG. 5 is a schematic diagram of the improved Xception network in the Deeplabv3+ model in the method of the present invention.

FIG. 6 is a schematic diagram of the implementation effect of the method of the present invention.

Detailed ways

1 is a schematic flow chart of the method of the method of the present invention: this high-definition aerial survey image-based power transmission and transformation engineering graveyard identification method provided by the present invention includes the following steps:

S1. Obtain remote sensing image data;

S2. Preprocessing the remote sensing image data obtained in step S1 to obtain training data; specifically, performing data enhancement on the remote sensing image data obtained in step S1 and expanding sample data to obtain training data;

During specific implementation, operations such as color enhancement and geometric transformation can be performed to expand the amount of sample data;

S3. Build a neural network model for cemetery recognition; specifically, build a deep learning framework TensorFlow to build a neural network model for cemetery object recognition; and use the Deeplabv3+ model as the neural network model;

In the specific implementation, the model framework adopts the Deeplabv3+ model framework; the Backbone of the model adopts Xception; ASPP is used for feature extraction; bilinear interpolation is used for upsampling for sampling; sliding window is used for prediction;

The overall architecture of the DeepLabv3+ model is shown in Figure 2: the main body of the Encoder is a DCNN with atrous convolution. Among them, DCNN can use a commonly used classification network such as ResNet, and then a spatial pyramid pooling module with atrous convolution (Atrous Spatial Pyramid Pooling, ASPP)), mainly to introduce multi-scale information; compared with DeepLabv3, v3+ introduced Decoder module, which further fuses low-level features with high-level features to improve the accuracy of segmentation boundaries.

In DeepLab, the ratio of the scale of the input image to the output feature map is recorded as output_stride. The DCNN here can be any classification network, generally called backbone, such as the ResNet network.

Atrous Convolution is one of the keys of the DeepLab model, which can control the receptive field without changing the size of the feature map, which is beneficial to extract multi-scale information. The atrous convolution is shown in Figure 3, where rate(r) controls the size of the receptive field, and the larger the r, the larger the receptive field.

In DeepLab, the Spatial Pyramid Pooling (ASPP) module is used to further extract multi-scale information, and here atrous convolutions with different rates are used to achieve this. ASPP is mainly to capture multi-scale information, which is crucial for segmentation accuracy. The ASPP module mainly includes the following parts:

(1) A 1×1 convolutional layer, and three 3x3 hole convolutions, for output_stride=16, the rate is (6, 12, 18), if output_stride=8, the rate is doubled (the output of these convolutional layers is The number of channels is 256 and contains BN layer);

(2) A global average pooling layer obtains image-level features, which are then sent to a 1x1 convolutional layer (outputting 256 channels), and bilinearly interpolated to the original size;

(3) Concat the four features of different scales obtained in part (1) and part (2) together in the channel dimension, and then send it into a 1x1 convolution for fusion and obtain a new feature of 256-channel.

For DeepLabv3, the output_stride of the feature map obtained by the ASPP module is 8 or 16, which is directly bilinearly interpolated to the original image size after passing through the 1x1 classification layer. This is a very violent decoder method, especially output_stride=16. However, this is not conducive to obtaining finer segmentation results, so the DeepLabv3+ model draws on the EncoderDecoder structure and introduces a new Decoder module, as shown in Figure 4. First, bilinearly interpolate the features obtained by the encoder to obtain 4x features, and then concat with the low-level features of the corresponding size in the encoder, such as the Conv2 layer in ResNet, because the number of features obtained by the encoder is only 256, and the low-level feature dimension may be very high , in order to prevent the high-level features obtained by the encoder from being weakened, 1x1 convolution is used to reduce the dimension of the low-level features (the output dimension in the paper is 48). After the two features are concat, 3x3 convolution is used to further fuse the features, and finally bilinear interpolation is used to obtain a segmentation prediction of the same size as the original image.

The backbone used by DeepLabv3 is the ResNet network, and an improved Xception is tried in the v3+ model. The Xception network mainly uses Depthwise Separable Convolution, which makes Xception less computationally intensive (as shown in Figure 5). The improved Xception is mainly reflected in the following points:

1. Referring to the modification of MSRA, more layers have been added;

2. All max pooling layers are replaced with depthwise separable convolutions with stride=2, which can be changed to hole convolution;

3. Add BN and ReLU after 3x3 depthwise convolution.

Using the improved Xception network as the backbone, the DeepLab network segmentation effect has been improved to a certain extent.

S4. using the training data obtained in step S2 to train the cemetery identification neural network model constructed in step S3, thereby obtaining a cemetery identification model;

S5. Use the cemetery identification model obtained in step S4 to identify the remote sensing image data to be identified, thereby obtaining an identification result;

S6. Perform optimization processing on the identification result obtained in step S5, thereby obtaining the final burial ground identification result; in specific implementation, the optimization processing includes noise point removal processing, smoothing processing, and breakpoint connection processing.

In addition, after the model is trained, model compression technology can be used to compress the established recognition model, thereby improving the efficiency of the model; specifically, distillation, pruning and other methods can be used to compress the model.

FIG. 6 is a schematic diagram of a specific implementation of the method of the present invention. Fig. 6(a) and Fig. 6(b) are a pair; Fig. 6(a) is the original unrecognized image, and in Fig. 6(b), in the upper left part, the identified graveyard is marked with a solid area Fig. 6(c) and Fig. 6(d) are a pair; Fig. 6(d) is the original unrecognized image, and the identified burial area is marked with a solid area in Fig. 6(d). It can be seen from FIG. 6 that the implementation effect of the method of the present invention is good, and the recognition accuracy rate is high.

Claims (6)

1. A power transmission and transformation engineering graveyard identification method based on high-definition aerial survey images, comprising the steps: S1. Obtain remote sensing image data; S2. Preprocess the remote sensing image data obtained in step S1 to obtain training data; S3. Build a neural network model for cemetery identification; S4. using the training data obtained in step S2 to train the cemetery identification neural network model constructed in step S3, thereby obtaining a cemetery identification model; S5. Use the cemetery identification model obtained in step S4 to identify the remote sensing image data to be identified, thereby obtaining an identification result; S6. Optimizing the identification result obtained in step S5, so as to obtain the final identification result of the cemetery. 2. The power transmission and transformation engineering graveyard identification method based on high-definition aerial survey images according to claim 1, is characterized in that the remote sensing image data obtained in step S1 is preprocessed according to step S2, thereby obtaining training data, specifically for The remote sensing image data obtained in step S1 is subjected to data enhancement and sample data expansion to obtain training data. 3. the power transmission and transformation engineering cemetery identification method based on high-definition aerial survey image according to claim 2, it is characterized in that the described building burial ground identification neural network model described in step S3, is specifically to build deep learning framework TensorFlow, thereby build cemetery objects A neural network model for recognition. 4. The power transmission and transformation engineering graveyard identification method based on high-definition aerial survey images according to claim 3, wherein the neural network model is a Deeplabv3+ model. 5 . The method for identifying a power transmission and transformation engineering graveyard based on high-definition aerial survey images according to claim 4 , wherein the optimization processing in step S6 specifically includes noise point removal processing, smoothing processing and breakpoint connection processing. 6 . 6. The method for identifying a power transmission and transformation project cemetery based on high-definition aerial survey images according to one of claims 1 to 5, characterized in that it further comprises the following steps: Model compression technology is used to compress the established recognition model, so as to improve the efficiency of the model.

Images