New energy station key component inspection and temperature measurement method and system based on deep learning

Abstract

The invention relates to the technical field of target detection and temperature detection of a new energy station, in particular to a new energy station key component inspection and temperature measurement method and system based on deep learning, and the method comprises the steps: collecting infrared thermal images of a cable, a combiner box, a connecting piece and other target pieces of the new energy station through a thermal imager; after preprocessing, data labeling is carried out, classification labels are made, and an infrared thermogram data set is obtained; an improved YOLOv5 deep learning algorithm training model is established, a GhostNet module is used to reduce the calculation amount, a CBAM module is used to improve the feature extraction capability, the features of the target component of the new energy station are extracted, the model is trained according to the obtained data set, the target component is identified, and the temperature of the target component is detected. According to the invention, the temperature anomaly diagnosis of the key equipment of the new energy station can be accurately and efficiently carried out in real time.

Description

Technical Field

[0001] The present invention relates to the technical fields of target detection and temperature detection in new energy power stations, and particularly to a method and system for inspecting and measuring the temperature of key components in a new energy power station based on deep learning. Background Art

[0002] The good operating state of key components such as cables, busbar trunking units, and connectors in new energy power stations is the guarantee for the safe operation of energy. With the increasingly intensive and heavy use of new energy, ensuring the safe and stable operation of key equipment in a high-speed and intensive operating environment is the key to the sustainable development of new energy power stations. By accurately monitoring and predicting the temperature of key components, potential fault signs can be detected in a timely manner, the health status of the equipment can be improved, the service life can be extended, the maintenance cost can be reduced, and the safety and reliability of the power station operation can be guaranteed. Therefore, researching reliable detection methods and establishing reasonable temperature inspection models for key components such as cables, busbar trunking units, and connectors in new energy power stations is of great significance for ensuring the safety of new energy power stations and reducing maintenance costs.

[0003] Currently, for temperature measurement of key equipment in new energy power stations in China, most use two temperature detection systems: direct contact type and non-direct contact type. The maintenance and calibration workload of contact sensors in the contact temperature measurement technology under harsh environments is very cumbersome, which greatly increases the operation cost of new energy power stations. Moreover, a single contact temperature measurement point on each device poses a great risk of misjudgment for the temperature detection system. The current non-direct contact temperature detection system mainly uses infrared single-point detection technology. By measuring the temperature infrared rays emitted by the infrared sensor, the temperature value of a certain point of the key component is measured. Such a temperature measurement method can only obtain the local temperature of the key component and cannot fully reflect the heat generation situation of the entire component. For the position of the key component identified by the infrared thermal image, there are currently methods of using traditional manually designed feature extractors and methods of using deep learning neural networks, but the accuracy is relatively low. Summary of the Invention

[0004] The present invention provides a method and system for inspecting and measuring the temperature of key components in a new energy power station based on deep learning, which can accurately, efficiently, and real-time diagnose high-temperature faults of key equipment in a new energy power station.

[0005] In order to achieve the object of the present invention, the technical solution adopted is: A method for inspecting and measuring the temperature of key components in a new energy power station based on deep learning, including the following steps:

[0006] S1. Set up an infrared image acquisition system for the new energy power station, and collect infrared thermal images of key target components in the new energy power station through the infrared image acquisition system;

[0007] S2. Preprocess the infrared thermal image by adding noise, rotating, and scaling, and perform data annotation on the processed infrared thermal image to create classification labels and temperature labels for the target components, obtaining an infrared thermal image dataset;

[0008] S3. Establish an improved YOLOv5 deep learning model, use the GhostNet module to replace the original convolutional module, integrate the convolutional attention module CBAM into the YOLOv5 backbone network, use CIoU as the bounding box regression loss function for the target, extract the features of the target components in the new energy station, and use the infrared thermal image dataset to train the target detection model to identify the information of the target components in the new energy station;

[0009] S4. Establish a temperature detection deep learning model based on MobileNet, use the mean squared error loss function, and train the temperature detection deep learning model according to the pixel information of the target components in the infrared thermal image and the temperature labels of the target components obtained in step S2;

[0010] S5. Use the target detection model and the temperature detection deep learning model trained in step S3 and step S4 to establish an integrated inspection and temperature measurement module that returns results in real time.

[0011] As an optimized solution of the present invention, in step S1, the infrared image acquisition system includes an infrared thermal imager and a temperature sensor. The infrared thermal imager is installed within a distance of 2 meters from the target component, and infrared thermal images of the required target components are collected from various angles at fixed time intervals. The temperature sensor is installed on the target component, and the temperature sensor synchronously collects the temperature data of the target component during the infrared thermal image acquisition. The infrared thermal imager performs the acquisition of pixel temperature points, and the temperature sensor real-time collects the temperature of the target components in the new energy station, obtaining the infrared thermal images and temperature data of the key target components.

[0012] As an optimized solution of the present invention, in step 2, it specifically includes the following steps:

[0013] S2-1. Perform preprocessing on the infrared thermal image collected by the infrared image acquisition system. The preprocessing includes adding noise, rotating, and scaling to achieve data augmentation;

[0014] S2-2. Use LabelImg software to perform data annotation on the infrared thermal image after data augmentation, mark the pixel coordinates of the target components in the image, generate a label file in xml format, and create classification labels and temperature labels for each target component in the image to complete the construction of the target dataset;

[0015] S2-3. When making classification labels using LableImg software, multiple categories are defined, that is, the categories in the voc_class.txt file; after data annotation, the.xml files with labels and the corresponding source images are saved in the dataset format of PASCAL VOC2007 to obtain an infrared thermal image dataset.

[0016] As an optimized solution of the present invention, in step S3, an improved YOLOv5 deep learning model is established, and the specific process is as follows:

[0017] S3-1. For the improved YOLOv5 deep learning model, the GhostNet module constructs a Ghost structure C3Ghost suitable for YOLOv5 by using GhostNet and combining the original CSPNet, and incorporates the convolutional module attention mechanism module CBAM into the backbone network of YOLOv5, and inserts the convolutional module attention mechanism module CBAM between the Backbone and Neck modules of YOLOv5;

[0018] S3-2. The CIoU loss function is used for model training. The CIoU loss function considers the overlapping area between the predicted box and the corresponding key target component, and also considers the distance between the center points and the aspect ratios of the two bounding boxes. The calculation formula is as follows:

[0019]

[0020]

[0021]

[0022] Among them, IoU represents the intersection over union, ρ represents the Euclidean distance between the predicted box and the true predicted box, b represents the center point of the true box, b gt represents the center point of the predicted box, c represents the shortest diagonal length of the smallest box containing the key target component and the predicted box, α is a weight parameter, ν represents the similarity between the aspect ratios of the two bounding boxes, w gt and h gt represent the width and height of the key target component, w and h represent the width and height of the predicted box respectively, and CIoU represents the complete bounding box regression loss;

[0023] S3-3. Set the training parameters of the object detection model.

[0024] As an optimized solution of the present invention, in step S4, a deep learning model for detecting temperature is established, and the specific process is as follows:

[0025] S4-1. Construct a deep learning convolutional model based on MobileNet, and use a regression layer in the last layer to predict the temperature;

[0026] S4-2. Use the mean squared error loss function, and train the model with the pixel values of the target components in the image dataset obtained in step S2 and the corresponding temperature labels, so as to obtain a temperature detection deep learning model that can predict the temperature using the pixel values of the infrared thermal image;

[0027] S4-3. Set the training parameters of the temperature detection deep learning model.

[0028] As an optimized solution of the present invention, in step S5, a comprehensive inspection and temperature measurement module that returns results in real time is established, which combines the results returned by the target detection model and the temperature detection deep learning model to generate real-time image information, mark the target category and predicted temperature, and display the results.

[0029] A key component inspection and temperature measurement system for a new energy power station based on deep learning, characterized by comprising an infrared image acquisition module, an infrared thermal image processing module, a target recognition model module, a temperature detection model module, and a real-time result display module;

[0030] The infrared image acquisition module acquires infrared thermal images of target components in the new energy power station;

[0031] The infrared thermal image processing module is used to perform preprocessing on the infrared thermal images obtained in the infrared image acquisition module, including noise addition, rotation, and scaling, and perform data annotation on the preprocessed infrared thermal images to make classification labels, so as to obtain an infrared thermal image dataset;

[0032] The target detection module is used to establish an improved YOLOv5 deep learning model, integrate GhostNet and the convolutional attention module CBMA into the YOLOv5 backbone network, use CIoU as the bounding box regression loss function for the target, extract the features of the target components, and train the model according to the dataset obtained by the infrared thermal image processing module to identify the information of the target components in the new energy power station;

[0033] The temperature detection module is used to establish a temperature detection deep learning model based on the MobileNet convolutional network and a regression layer, use the mean squared error loss function, and train the temperature detection deep learning model with the pixel information of the target components in the infrared thermal image and the temperature labels of the target components to predict the pixel temperature of the identified target components;

[0034] The real-time result display module uses the P20Max embedded thermal imager to collect real-time infrared thermal images, and transmits the collected results to the processing program in the smart glasses for detection. This processing program uses the target detection module to complete target recognition, uses the temperature detection module to complete temperature prediction, marks the corresponding target component category information and temperature information on the infrared thermal image, and then displays the results in real time on the smart glasses.

[0035] An intelligent glasses mobile terminal includes a display, a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the above-mentioned key component inspection and temperature measurement method for new energy power stations based on deep learning.

[0036] A computer-readable storage medium stores a computer program thereon. When the program is executed by a processor, it implements the steps in the above-mentioned key component inspection and temperature measurement method for new energy power stations based on deep learning.

[0037] The present invention has positive effects: 1) The present invention classifies and identifies key equipment based on the infrared thermal images of new energy power stations, constructs a dataset using on-site infrared thermal images and trains a model. Compared with the traditional YOLOv5 algorithm, replacing the original convolution module with GhostNet improves the calculation speed, and integrating the CBAM attention mechanism into the original model further extracts the features of the target, further improving the detection accuracy;

[0038] 2) The present invention uses CIoU as the bounding box regression loss function for the target, making the prediction of the target position more accurate;

[0039] 3) The present invention establishes a temperature detection deep learning model using MobileNet, constructs a dataset using on-site infrared thermal images and trains the model, with a higher accuracy rate compared to traditional temperature calculation methods;

[0040] 4) The present invention uses a result return program to display the detection results in real time, enabling convenient, real-time, and wearable temperature inspection. Description of the Drawings

[0041] The following further elaborates on the present invention in detail in conjunction with the drawings and specific embodiments.

[0042] Figure 1 is the flowchart of the present invention;

[0043] Figure 2 is the network structure diagram of the improved algorithm based on YOLOv5 of the present invention;

[0044] Figure 3 is the specific network structure diagram of the attention mechanism of the present invention.

[0045] Figure 4 This is the result graph of the improved model object detection in the embodiments of the present invention.

[0046] Figure 5 This is the PR curve graph of the model training of the present invention.

[0047] Figure 6 This is the schematic diagram of the temperature detection of key components in a new energy power station based on deep learning in the embodiments of the present invention. Detailed implementation manners

[0048] As Figure 1 shown, the present invention discloses a method for inspecting and measuring the temperature of key components in a new energy power station based on deep learning, including the following steps:

[0049] S1. Set up an infrared image acquisition system for the new energy power station, and collect infrared thermal images of key target components in the new energy power station through the infrared image acquisition system; the key target components include cables, busbar boxes, connectors, etc.

[0050] In step S1, the infrared image acquisition system includes an infrared thermal imager and a temperature sensor. The infrared thermal imager is installed within a distance of 2 meters from the target component, and infrared thermal images of the required target components are collected from various angles at fixed time intervals. The temperature sensor is installed on the target component, and the temperature sensor synchronously collects the temperature data of the target component during the infrared thermal image acquisition. The infrared thermal imager performs the acquisition of pixel temperature points, and the temperature sensor real-time collects the temperature of the target components in the new energy power station, obtaining the infrared thermal images and temperature data of the key target components

[0051] S2. Perform preprocessing on the infrared thermal images, including adding noise, rotation, and scaling, and perform data annotation on the processed infrared thermal images, making classification labels and temperature labels for the target components, to obtain an infrared thermal image data set; the specific process is as follows:

[0052] S2-1. For the infrared thermal images collected by the infrared image acquisition system in step S1, perform preprocessing, and the preprocessing includes adding noise, rotation, and scaling processing to achieve data enhancement;

[0053] S2-2. Use LabelImg software to perform data annotation on the infrared thermal images after data augmentation, mark the pixel coordinates of the target components in the images, generate a lable file in xml format, make classification labels and temperature labels for each target component in the images, and complete the construction of the target data set;

[0054] S2-3. When making classification labels using LableImg software, multiple categories are defined, namely the categories in the voc_class.txt file. After data annotation, the.xml files with labels and the corresponding source images are saved in the dataset format of PASCAL VOC2007 to obtain the infrared thermal image dataset.

[0055] S3. Establish an improved YOLOv5 deep learning model. Use the GhostNet module to replace the original convolution module to reduce the computational load, and integrate the convolutional attention module CBAM into the YOLOv5 backbone network. Use CIoU as the bounding box regression loss function for the target, extract the features of the target components in the new energy station yard, and use the infrared thermal image dataset to train the target detection model to identify the target component information of the new energy station yard.

[0056] The specific process is as follows:

[0057] S3-1. For the improved YOLOv5 deep learning model, the GhostNet module constructs the Ghost structure C3Ghost suitable for YOLOv5 by leveraging the advantages of GhostNet and combining the original CSPNet, and incorporates the attention mechanism module CBAM of the convolutional module into the backbone network of YOLOv5, and inserts the attention mechanism module CBAM of the convolutional module between the Backbone and Neck modules of YOLOv5.

[0058] S3-2. Use the CIoU loss function for model training. The CIoU loss function considers the overlapping area between the predicted bounding box and the corresponding key target component, and also considers the distance between the center points and the aspect ratios of the two bounding boxes. The calculation formula is as follows:

[0059]

[0060]

[0061]

[0062] Among them, IoU represents the intersection over union, ρ represents the Euclidean distance between the predicted bounding box and the ground truth bounding box, b represents the center point of the ground truth box, b gt represents the center point of the predicted bounding box, c represents the shortest diagonal length of the smallest box containing the key target component and the predicted bounding box, α is the weight parameter, ν represents the similarity between the aspect ratios of the two bounding boxes, wg and h gt represent the width and height of the key target component, w and h represent the width and height of the predicted bounding box respectively, and CIoU represents the complete bounding box regression loss.

[0063] S3-3. Set the training parameters of the target detection model. The parameter settings are as follows: classes = 3 in the [yolo] section of the yolov5.cfg file; set the number of iterations in the training code train.py, that is, set the parameter epochs; save the generated weights model.save_weights(log_dir + 'train_weight.hs'), where model.save_weights represents saving the weights of the model, log_dir represents the saving path, and train_weight.hs represents the training weights.

[0064] S4. Establish a deep learning model for temperature detection based on MobileNet, use the mean squared error loss function, and train the temperature detection deep learning model according to the pixel information of the target components in the infrared thermal image and the temperature labels of the target components obtained in step S2. The specific process is as follows:

[0065] S4-1. Build a deep learning convolutional model based on MobileNet, and use a regression layer to predict the temperature in the last layer.

[0066] S4-2. Use the mean squared error loss function to train the model with the pixel values of the target components in the image dataset and the corresponding temperature labels obtained in step S2, so as to obtain a temperature detection deep learning model that can predict the temperature using the pixel values of the infrared thermal image.

[0067] S4-3. Set the training parameters of the temperature detection deep learning model

[0068] S5. Use the target detection model and the temperature detection deep learning model trained in steps S3 and S4 to establish a comprehensive inspection and temperature measurement module that returns results in real time.

[0069] In step S5, establish a comprehensive inspection and temperature measurement module that returns results in real time, combine the results returned by the target detection model and the temperature detection deep learning model, generate real-time image information, mark the target category and predicted temperature, and display the results.

[0070] As Figure 6 shown, a key component inspection and temperature measurement system for new energy power stations based on deep learning includes an infrared image acquisition module, an infrared thermal image processing module, a target recognition model module, a temperature detection model module, and a real-time result display module;

[0071] The infrared image acquisition module acquires the infrared thermal images of the target components in the new energy power station;

[0072] An infrared thermal image processing module is used to preprocess the infrared thermal images obtained by the infrared image acquisition module, including adding noise, rotating, and scaling, and to perform data annotation on the preprocessed infrared thermal images, create classification labels, and obtain an infrared thermal image dataset;

[0073] A target detection module is used to establish an improved YOLOv5 deep learning model, integrate GhostNet and the convolutional attention module CBMA into the YOLOv5 backbone network, use CIoU as the bounding box regression loss function for the target, extract target component features, and train the model based on the dataset obtained by the infrared thermal image processing module to identify the information of target components in new energy power stations;

[0074] A temperature detection module is used to establish a temperature detection deep learning model based on the MobileNet convolutional network and the regression layer, use the mean squared error loss function, and train the temperature detection deep learning model with the pixel information of the target components in the infrared thermal images and the temperature labels of the target components to predict the pixel temperature of the identified target components;

[0075] A real-time result display module uses a P20Max embedded thermal imager to collect real-time infrared thermal images, transmits the collected results to the processing program in the smart glasses for detection. This processing program uses the target detection module to complete target recognition, uses the temperature detection module to complete temperature prediction, annotates the corresponding target component category information and temperature information on the infrared thermal images, and then displays the results in real-time on the smart glasses.

[0076] A smart glasses mobile terminal includes a display, a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the above-mentioned method for inspecting and measuring the temperature of key components in new energy power stations based on deep learning.

[0077] A computer-readable storage medium stores a computer program, and when the program is executed by a processor, it implements the steps in the above-mentioned method for inspecting and measuring the temperature of key components in new energy power stations based on deep learning.

[0078] The following combines Figures 2 to 6 and specific implementation examples to further illustrate the present invention.

[0079] Embodiment

[0080] The method for inspecting and measuring the temperature of key components in new energy power stations based on deep learning of the present invention uses an improved YOLOv5 deep learning algorithm that integrates GhostNet and the attention mechanism. The improved network structure model is as Figure 2 , and the specific structure of the attention mechanism module is as Figure 3 .

[0081] Step 1: Set up an infrared image acquisition system for the new energy power station, and collect infrared thermal images of key target components such as cables, busbar boxes, and connectors in the new energy power station through the infrared image acquisition system.

[0082] The specific operation of Step 1 is to use an infrared thermal imager and a temperature sensor. The infrared thermal imager is installed within a distance of 2 meters from the target component, and infrared thermal images of the required target components are collected from different angles at fixed intervals. The temperature sensor is installed on the target component to synchronously collect the temperature data of the target component during the infrared thermal image acquisition.

[0083] Step 2: After obtaining the infrared images, it is necessary to perform data annotation and label production on the images.

[0084] Further, the specific operation of Step 2 is as follows:

[0085] (1) Perform image preprocessing on the collected infrared thermal images, including adding noise, rotating, and scaling for image enhancement.

[0086] (2) Use the LabelImg software for data annotation to generate the corresponding xml-format label file; and make the corresponding classification labels and temperature labels; finally, complete the construction of the target dataset.

[0087] (3) After data annotation, save the.xml file with labels and its corresponding source image in the dataset format of PASCAL_VOC2007. Among them, the data folder data is established in the same directory as the training code train.py.

[0088] Step 3: Improve YOLOv5, utilize the advantages of GhostNet, and combine with the original CSPNet to design and construct a lightweight Ghost structure C3Ghost suitable for YOLOv5, and integrate the convolutional attention module into the YOLOv5 backbone network to improve its feature extraction ability. Use CIoU as the bounding box regression loss function for the target.

[0089] Use the CIoU loss function for model training. The CIoU loss function considers the overlapping area between the predicted box and the corresponding key target component, and also considers the distance between the center points and the aspect ratios of the two bounding boxes. The calculation formula is as follows:

[0090]

[0091]

[0092]

[0093] Among them, IoU represents the intersection over union, ρ represents the Euclidean distance between the predicted bounding box and the ground truth bounding box, b represents the center point of the ground truth box, and b gt represents the center point of the predicted bounding box, c represents the length of the shortest diagonal of the smallest box containing the key target component and the predicted bounding box, α is the weight parameter, ν represents the similarity between the aspect ratios of the two bounding boxes, w gt and h gt represent the width and height of the key target component, w and h represent the width and height of the predicted bounding box respectively, and CIoU represents the complete bounding box regression loss.

[0094] Step 4: According to the constructed training set and validation set, train the model and verify its performance.

[0095] Step 5: Use the best.py weight file after model training to detect the new energy power station images and output the detection results. The relevant parameter settings for model training are as follows: classes = 3 in the [yolo] section of the yolov5.cfg file; the number of iterations can be changed in the training code train.py, that is, set the parameter epochs; save the generated weights: model.save_weights(log_dir + 'train_weight.hs'), where model.save_weights represents saving the model weights, log_dir represents the saving path, and train_weight.hs represents the training weights. The results of target classification prediction and recognition are as Figure 4 . The PR curve of the results is as Figure 5 shown, and the following table can be obtained.

[0096] Comparison table of improved YOLOv5 and original YOLOv5 results

[0097] Algorithm model Training duration / h FPS mAP(%) mAP@0.5:0.95(%) YOLOv5s 4.5 61 87.8 85.6 Improved YOLOv5s 4.1 68.5 96.5 94.3

[0098] As can be seen from the above table, when the improved model shows similar performance in terms of training time and training time per single image, the average prediction accuracy can be increased by 8.7% compared to the original model, demonstrating the effectiveness of the improved model.

[0099] In Step 6, a deep learning model for detecting temperature is established, and the specific process is as follows:

[0100] 4.1 Construct a deep learning convolutional model based on MobileNet, and use a regression layer to predict the temperature in the last layer;

[0101] 4.2 Use the mean squared error loss function, and train the model using the pixel values of the target components in the image dataset obtained in Step 2 and the corresponding temperature labels, so as to obtain a deep learning model that can predict the temperature using the pixel values of the infrared thermal image.

[0102] Step 7: Train the temperature model based on the constructed training set and validation set and verify its performance.

[0103] Step 8: Use the result return integration module to generate a target image by combining the results returned by the target detection model and the temperature detection model, and label the target category and predicted temperature.

[0104] It should be understood that, in order to streamline the present invention and assist those skilled in the art in understanding various aspects of the present invention, in the above description of the exemplary embodiments of the present invention, various features of the present invention are sometimes described in a single embodiment or with reference to a single figure. However, the present invention should not be construed as meaning that the features included in the exemplary embodiments are all essential technical features of the patent claims of the present invention.

[0105] The specific embodiments described above have further elaborated on the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for temperature detection of key components in new energy power stations based on deep learning, characterized in that: It includes the following steps: S1. Set up an infrared image acquisition system for the new energy power station, and collect infrared thermal images of key target components in the new energy power station through the infrared image acquisition system; S2. Perform preprocessing on the infrared thermal images, including adding noise, rotation, and scaling, and perform data annotation on the processed infrared thermal images to make classification labels and temperature labels for the target components, obtaining an infrared thermal image dataset; S3. Establish an improved YOLOv5 deep learning model, use the GhostNet module to replace the original convolution module, integrate the convolutional attention module CBAM into the YOLOv5 backbone network, use CIoU as the bounding box regression loss function for the target, extract the features of the target components in the new energy power station, and use the infrared thermal image dataset to train the target detection model to identify the target component information in the new energy power station; S4. Establish a temperature detection deep learning model based on MobileNet, use the mean squared error loss function, and train the temperature detection deep learning model according to the pixel information of the target components in the infrared thermal images obtained in step S2 and the temperature labels of the target components; S5. Use the target detection model and the temperature detection deep learning model trained in step S3 and step S4 to establish an integrated inspection and temperature measurement module that returns results in real time.

2. The key component inspection and temperature measurement method for new energy power stations based on deep learning according to claim 1, characterized in that: In step S1, the infrared image acquisition system includes an infrared thermal imager and a temperature sensor. The infrared thermal imager is installed within a distance of 2 meters from the target component, and infrared thermal images of the required target components are collected from various angles at fixed time intervals. The temperature sensor is installed on the target component, and the temperature sensor synchronously collects the temperature data of the target component during the infrared thermal image acquisition. The infrared thermal imager collects pixel temperature points, and the temperature sensor real-time collects the temperature of the target components in the new energy power station, obtaining the infrared thermal images and temperature data of the key target components.

3. The method for detecting and measuring the temperature of key components in a new energy power station based on deep learning according to claim 2, wherein: In step 2, it specifically includes the following steps: S2-1. For the infrared thermal images collected by the infrared image acquisition system, perform preprocessing, including adding noise, rotation, and scaling, to achieve data augmentation; S2-2. Use the LabelImg software to perform data annotation on the infrared thermal images after data augmentation, mark the pixel coordinates of the target components in the images, generate lable files in xml format, make classification labels and temperature labels for each target component in the images, and complete the construction of the target dataset; S2-3. When using the LabelImg software to make classification labels, multiple categories are defined, that is, the categories in the voc_class.txt file; after data annotation, save the.xml files with labels and the corresponding source images in the dataset format of PASCAL VOC2007 to obtain the infrared thermal image dataset.

4. The method for detecting and measuring the temperature of key components in a new energy power station based on deep learning according to claim 3, characterized in that: In step S3, to establish an improved YOLOv5 deep learning model, the specific process is as follows: S3-1. An improved YOLOv5 deep learning model. The GhostNet module constructs a Ghost structure C3Ghost suitable for YOLOv5 by utilizing GhostNet and combining it with the original CSPNet. An attention mechanism module CBAM for convolutional modules is incorporated into the backbone network of YOLOv5, and the attention mechanism module CBAM for convolutional modules is inserted between the Backbone and Neck modules of YOLOv5. S3-2. Use the CIoU loss function for model training. The CIoU loss function takes into account the overlapping area between the predicted bounding box and the corresponding key target components, as well as the distance between the center points and the aspect ratios of the two bounding boxes. The calculation formula is as follows: Among them, IoU represents the intersection over union, ρ represents the Euclidean distance between the predicted bounding box and the ground-truth bounding box, b represents the center point of the ground-truth bounding box, b gt represents the center point of the predicted bounding box, c represents the length of the shortest diagonal of the smallest bounding box containing the key target component and the predicted bounding box, α is the weight parameter, ν represents the similarity between the aspect ratios of the two bounding boxes, w gt and h gt represent the width and height of the key target component, w and h respectively represent the width and height of the predicted bounding box, and CIoU represents the complete bounding box regression loss; S3-3. Set the training parameters of the object detection model.

5. The method for detecting and temperature measuring of key components in a new energy power station based on deep learning according to claim 4, wherein: In step S4, a deep learning model for detecting temperature is established. The specific process is as follows: S4-1. Construct a deep learning convolutional model based on MobileNet, and use a regression layer in the last layer to predict the temperature. S4-2. Use the mean squared error loss function, and train the model with the pixel values of the target components in the image dataset obtained in step S2 and the corresponding temperature labels, so as to obtain a temperature detection deep learning model that can predict the temperature using the pixel values of the infrared thermal image. S4-3. Set the training parameters of the temperature detection deep learning model.

6. The method for detecting and measuring the temperature of key components in a new energy power station based on deep learning according to claim 5, wherein: In step S5, a comprehensive temperature detection module for real-time result return is established. Combining the results returned by the object detection model and the temperature detection deep learning model, real-time image information is generated, the target category and predicted temperature are marked, and the results are displayed.

7. A key component inspection and temperature measurement system for new energy power stations based on deep learning, characterized in that, It includes an infrared image acquisition module, an infrared thermal image processing module, an object recognition model module, a temperature detection model module, and a real-time result display module. The infrared image acquisition module acquires the infrared thermal image of the target components in the new energy power station. The infrared thermal image processing module is used to perform preprocessing on the infrared thermal image obtained in the infrared image acquisition module, including noise addition, rotation, and scaling, and perform data annotation on the preprocessed infrared thermal image to make classification labels, obtaining an infrared thermal image dataset. The object detection module is used to establish an improved YOLOv5 deep learning model, integrate GhostNet and the convolutional attention module CBMA into the YOLOv5 backbone network, use CIoU as the bounding box regression loss function for the object, extract the features of the target components, and train the model according to the dataset obtained by the infrared thermal image processing module to identify the information of the target components in the new energy power station. The temperature detection module is used to establish a temperature detection deep learning model based on the MobileNet convolutional network and the regression layer, use the mean squared error loss function, and train the temperature detection deep learning model with the pixel information of the target components in the infrared thermal image and the temperature labels of the target components to predict the pixel temperature of the identified target components. The real-time result display module uses an embedded thermal imager to collect real-time infrared thermal images, and transmits the collected results to the processing program in the smart glasses for detection. The processing program uses the target detection module to complete target recognition, uses the temperature detection module to complete temperature prediction, and marks the corresponding target component category information and temperature information on the infrared thermal image, and then displays the results on the smart glasses in real time.

8. An intelligent glasses mobile terminal, comprising a display, a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for inspecting and measuring the temperature of key components in a new energy power station based on deep learning according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the method for inspecting and measuring the temperature of key components in a new energy power station based on deep learning according to any one of claims 1 to 6.

Images