Intelligent detection system for sand spreading status of locomotives in maintenance yard

Abstract

The present invention relates to the field of machine learning and computer vision technology, and in particular to an intelligent detection system for the sand spreading state of a locomotive in a maintenance yard, comprising a camera, a license plate recognition device, an LED display screen, an IP audio system, a switch, an optical fiber transceiver, a hard disk recorder, and an intelligent analysis server. The intelligent analysis server trains the YOLOV5 convolutional neural network provided in the intelligent analysis server through the image information collected by the camera and the license plate recognition device, and uses the trained convolutional neural network to identify the degree of sand spreading of the locomotive to be detected. The convolutional neural network can learn the feature representation of sand spreading images in different states, so as to accurately classify and identify the degree of sand spreading of the locomotive, and the system equipment only needs to be installed at a fixed position in the maintenance yard, and does not need to be installed for each locomotive, thereby reducing hardware equipment and installation costs.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning and computer vision, and in particular to an intelligent detection system for the sand spreading state of a locomotive in a maintenance yard. Background Art

[0002] With the development of railway transportation, the operation safety of heavy-duty trucks on the line is particularly important. One of the key points is to ensure that the sand-spreading pipes on the locomotive are in good working condition to ensure the safety and efficiency of railway transportation. Ensuring that the sand-spreading pipes on the locomotive work properly is crucial to the safety and efficiency of railway transportation.

[0003] The patent with publication number CN110126853A discloses an intelligent monitoring and diagnosis system for sand spreading of railway locomotives, including a human-machine interaction system, a sand spreading signal collector, a photoelectric conversion module, a monitoring module, and a monitoring probe. The human-machine interaction system includes an integrated machine installed inside the sand spreading vehicle, a control panel is installed on the front outer surface of the integrated machine, the rear outer surface of the control panel is fixedly connected to the front outer surface of the integrated machine, and the signal port of the integrated machine is connected to the signal port of the control panel. The advantages of the present invention are: the intelligent monitoring and diagnosis system for sand spreading of railway locomotives can realize real-time monitoring of the sand spreading status of each sand spreading port, save the locomotive preparation operation time and locomotive sand, and at the same time reduce the probability of polluting the preparation operation site, reduce the sand spreading detection and maintenance costs, save workers' time to maintain equipment, prevent failures, effectively reduce the rate of machine breakage and temporary repairs, reduce the labor intensity of workers, and improve the reliability of locomotives, thereby improving the attendance guarantee rate of locomotives and increasing transportation and efficiency.

[0004] It can be seen that the railway locomotive sand-spreading intelligent monitoring and diagnostic system has the following problems: the railway locomotive sand-spreading intelligent monitoring and diagnostic system has high costs due to on-board installation, and on-board installation also has the problems of increased installation complexity, difficult maintenance and poor compatibility. Summary of the invention

[0005] To this end, the present invention provides an intelligent detection system for the sand spreading status of a locomotive in a maintenance yard, so as to overcome the problems of increased installation complexity, difficult maintenance and poor compatibility in the prior art of on-vehicle installation.

[0006] To achieve the above object, the present invention provides an intelligent detection system for sand spreading status of locomotives in a maintenance yard, comprising:

[0007] Cameras are installed on both sides of the railway tracks in the maintenance yard to collect real-time images of locomotive sand spreading;

[0008] License plate recognition devices are installed on both sides of the railway track at the exit of the maintenance yard, with one device on each side to collect the license plate number information of the locomotive entering the station;

[0009] An LED display screen is arranged behind the license plate recognition device to prompt the locomotive to perform sand spreading work through image display;

[0010] An IP audio system is installed beside the track opposite to the LED display screen to prompt the locomotive to spread sand by voice;

[0011] The switches are respectively arranged in the front-end power distribution cabinet and the machine room cabinet. The switch arranged in the front-end power distribution cabinet is connected to the camera, the license plate recognition device, the LED display screen and the IP audio; the switch arranged in the machine room cabinet is respectively connected to the optical fiber transceiver for transmitting data in the optical fiber network by converting electrical signals and optical signals, the hard disk recorder for storing historical events, the sand-spreading client alarm terminal for alarming according to the sand-spreading working status, and the intelligent analysis server for processing the acquired electrical signals and analyzing the sand-spreading results, so as to realize the communication of all devices by building a local area network.

[0012] Furthermore, the intelligent analysis server uses the improved CSPDarknet as the backbone network for feature extraction, performs feature fusion through PANet, and uses the Anchor-based mechanism to generate prediction boxes and category probabilities in the detection head part of the YOLOV5 convolutional neural network.

[0013] Furthermore, the camera collects a number of locomotive sand-spreading images including sand-spreading and non-sand-spreading states, and at the same time, the license plate recognition device collects license plate number information and transmits it to the intelligent analysis server.

[0014] Furthermore, the intelligent analysis server compares the collected license plate number information with the locomotive models corresponding to all the license plate numbers stored therein, and identifies the locomotive model information corresponding to the collected locomotive sand-spreading images in the sand-spreading and non-sand-spreading states.

[0015] Furthermore, the intelligent analysis server determines a data set for the locomotive sand-spreading image to be input into the convolutional network neural network, and organizes it by category, wherein each category of images is stored in a corresponding folder and is divided into two parts: a training set and a validation set. The intelligent analysis server uses the albumentations library to perform random cropping, horizontal flipping, color jittering and rotation, adjusts the image to a specified input size of 224x224, and uses a bilinear interpolation algorithm to adapt the input layer of the model, normalizes the image pixel values, and scales the pixel values ​​to the range of [0,1].

[0016] Furthermore, the intelligent analysis server passes the preprocessed locomotive sand-spreading image as raw input to the convolutional neural network, the YOLOv5 convolutional neural network adopts CSPDarknet53 as its backbone network, and performs convolution operations in its convolutional layer. The YOLOv5 convolutional neural network calculates the local features of the image by weighted addition to obtain a feature map, uses a ReLU activation function to perform nonlinear mapping on the feature map, introduces residual connections, and adopts a CSP neural network structure. The feature map is divided into two parts for alternating updates. The convolutional neural network extracts image features related to the sand-spreading state from the original image through the stacking and optimization process of the entire convolutional layer and pooling layer.

[0017] Furthermore, the intelligent analysis server divides a part of the locomotive sand spreading state image data set as a validation set. After each training cycle, the validation set and the test set are used to calculate the training loss and the validation loss of the YOLOv5 convolutional neural network.

[0018] If the training loss is greater than the validation loss, the intelligent analysis server performs data regularization.

[0019] Furthermore, when the training loss is equal to the validation loss, the intelligent analysis server runs the model on the validation set and the test set respectively, calculates the prediction results of all samples, and calculates the accuracy of the validation set and the accuracy of the test set. If the accuracy of the test set is less than the accuracy of the validation set, the number of network layers of the YOLOv5 convolutional neural network is adjusted according to the difference between the accuracy of the validation set and the accuracy of the test set.

[0020] Furthermore, the intelligent analysis server performs the same preprocessing operation on the input new locomotive sand spreading image as the training data, inputs the preprocessed image into the trained convolutional neural network, extracts the image features layer by layer through the convolution layer, pooling layer and fully connected layer, generates the corresponding high-level representation, inputs the output of the last fully connected layer into the softmax function, generates the probability output of the three-classification task, and indicates the degree of locomotive sand spreading. The intelligent analysis server converts the output probability value into the classification result according to the preset range of the probability output preset in the server.

[0021] If the probability value is within the preset range, it is determined that the locomotive has an excellent sand spreading degree;

[0022] If the probability value is greater than or equal to the maximum value of the preset range, it is determined that the locomotive sand spreading degree is good;

[0023] If the probability is less than the minimum value of the preset range, it is determined that the locomotive has a poor degree of sand spreading.

[0024] Furthermore, the intelligent analysis server uses the Adam optimizer to adjust the model parameters to minimize the loss function and the model regularization to optimize and adjust the convolutional neural network parameters, and introduces the L1 norm penalty term through L1 regularization to sparse the parameter convolutional neural network parameters.

[0025] Compared with the prior art, the beneficial effect of the present invention is that the efficiency and accuracy of locomotive sand spreading operations are improved through automated image acquisition and analysis. Real-time monitoring and instant feedback mechanisms reduce human errors and enhance the safety of operations. The automated nature of the system also reduces labor costs, and optimizes the decision-making process through the deep learning model of the intelligent analysis server, thereby improving the intelligence level of the locomotive maintenance yard.

[0026] Furthermore, through this advanced network structure design, the intelligent analysis server can quickly and accurately detect the sand-spreading status of the locomotive, significantly improving the reliability and practicality of the detection system. The improved CSPDarknet53 backbone network reduces the amount of calculation while maintaining the quality of feature extraction. PANet's feature fusion technology enables the model to better understand the image content and improves the recognition accuracy of the sand-spreading status of different types of locomotives. The application of the Anchor-based mechanism enables the model to show higher accuracy and robustness in target detection tasks, thereby providing strong technical support for the maintenance and operation of locomotives in the maintenance yard.

[0027] Furthermore, by combining real-time image acquisition and license plate recognition technology, accurate monitoring and recording of the locomotive sand spreading status is achieved. This not only improves the efficiency and safety of locomotive maintenance, but also enhances the ability to identify locomotive status changes through the deep learning model of the intelligent analysis server, providing reliable data support for locomotive preventive maintenance and status assessment.

[0028] Furthermore, through the license plate number matching and vehicle model recognition functions of the intelligent analysis server, the system can realize personalized analysis and evaluation of the sand spreading status of locomotives of different models. This method improves the pertinence and accuracy of detection, makes maintenance work more precise, helps prevent maintenance deviations caused by vehicle model differences, and also provides data support for customized maintenance and services of locomotives, thereby improving the intelligence and automation level of maintenance yard operations.

[0029] Furthermore, through this preprocessing process of the intelligent analysis server, not only the generalization ability of the model for locomotive images in different sand spreading states is enhanced, the diversity of model training is increased through data enhancement, but also the standardization and consistency of image data are ensured, thereby improving the accuracy and robustness of the convolutional neural network in identifying the locomotive sand spreading state. In addition, the normalization process speeds up the model training speed and improves the stability of model convergence.

[0030] Furthermore, the recognition capability of the locomotive sand spreading status is enhanced through the CSPDarknet53 backbone network and PANet feature fusion technology. The use of ReLU activation function and residual connection improves the network's ability to express complex images and reduces the gradient vanishing problem during deep network training. In addition, through precise feature extraction, the system can more accurately evaluate the locomotive's sand spreading status, thereby providing reliable intelligent detection support for the maintenance yard.

[0031] Furthermore, when the training loss is lower than the validation loss, if this happens, the server will take data regularization measures, such as introducing L1 or L2 regularization, to reduce the overfitting of the model to the training data, thereby improving the generalization ability of the model. This ensures the stability and accuracy of the model in practical applications and improves the overall reliability of the detection system.

[0032] Furthermore, a detailed evaluation and adjustment mechanism allows the intelligent analysis server to accurately diagnose the performance of the model on different data sets and perform targeted structural optimization. By adjusting the number of network layers, the overfitting or underfitting problems that may exist in the model can be effectively solved, and the generalization ability of the model on unknown data can be improved, thereby ensuring the accuracy and reliability of the intelligent detection system in practical applications. In addition, this method also promotes the continuous improvement of model performance and adaptability.

[0033] Furthermore, through the classification method based on probability output and preset threshold, clear judgment criteria and operation procedures are provided for the intelligent detection of locomotive sand spreading status. It enables the system to automatically evaluate the locomotive sand spreading status and accurately classify it according to the probability output, thereby realizing real-time monitoring and evaluation of the locomotive sand spreading degree. This method not only improves the automation level of detection, but also enhances the interpretability of the system and the convenience of operation, which helps to take maintenance measures in time to ensure the safety and efficiency of railway transportation.

[0034] Furthermore, by using the Adam optimizer combined with the L1 regularization strategy, not only the efficiency and stability of model training are improved, but also the parameter sparsification brought by L1 regularization helps to reduce model complexity, prevent overfitting, and improve the generalization ability of the model. In addition, sparse parameters may also reduce the storage requirements and computing burden of the model, making the model more suitable for deployment in resource-constrained environments. The combined application of this optimization and regularization method improves the performance and reliability of the intelligent analysis server when processing locomotive sand spreading status detection tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a structural schematic diagram of the intelligent detection system for sand spreading status of locomotives in the maintenance yard of this embodiment;

[0036] Figure 2 This is a flow chart of data processing and model training performed by the intelligent analysis server in this embodiment;

[0037] Figure 3 A flow chart of model evaluation and adjustment performed by the intelligent analysis server of this embodiment;

[0038] Figure 4 This is a flow chart of intelligent detection and classification by the intelligent analysis server in this embodiment. DETAILED DESCRIPTION

[0039] In order to make the purpose and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0040] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0041] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0042] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0043] See also Figure 1-4 As shown, Figure 1 This is a structural schematic diagram of the intelligent detection system for sand spreading status of locomotives in the maintenance yard of this embodiment; Figure 2 This is a flow chart of data processing and model training performed by the intelligent analysis server in this embodiment;

[0044] Figure 3 A flow chart of model evaluation and adjustment performed by the intelligent analysis server of this embodiment; Figure 4 This is a flow chart of intelligent detection and classification by the intelligent analysis server in this embodiment.

[0045] This embodiment provides an intelligent detection system for sand spreading status of locomotives in a maintenance yard, including:

[0046] 180° wide-angle cameras are installed on both sides of the railway tracks in the maintenance yard to collect real-time images of locomotive sand spreading;

[0047] License plate recognition devices are installed on both sides of the railway track at the exit of the maintenance yard, with one device on each side to collect the license plate number information of the locomotive entering the station;

[0048] An LED display screen is arranged behind the license plate recognition device to prompt the locomotive to perform sand spreading work through image display;

[0049] An IP audio system is installed beside the track opposite to the LED display screen to prompt the locomotive to spread sand by voice;

[0050] The switches are respectively arranged in the front-end power distribution cabinet and the machine room cabinet. The switch arranged in the front-end power distribution cabinet is connected to the 180° wide-angle camera, the license plate recognition device, the LED display screen and the IP audio; the switch arranged in the machine room cabinet is respectively connected to the optical fiber transceiver for transmitting data in the optical fiber network by converting electrical signals and optical signals, the hard disk recorder for storing historical events, the sand-spreading client alarm terminal for alarming according to the sand-spreading working status, and the intelligent analysis server for processing the acquired electrical signals and analyzing the sand-spreading results, so as to realize the communication of all devices by building a local area network.

[0051] The 180° wide-angle cameras installed on both sides of the track collect the locomotive sand spreading images in real time, while the license plate recognition device captures the license plate information of the locomotive entering the station. The collected data is transmitted to the intelligent analysis server through the switch, which processes the data and analyzes the locomotive sand spreading status. The LED display and IP audio provide visual and auditory prompts to the locomotive driver based on the analysis results to ensure the correct execution of the sand spreading operation. The entire system realizes efficient communication between devices through the local area network.

[0052] The efficiency and accuracy of locomotive sand spreading operations are improved through automated image acquisition and analysis. Real-time monitoring and instant feedback mechanisms reduce human errors and enhance operational safety. The system’s automated features also reduce labor costs, and the deep learning model of the intelligent analysis server optimizes the decision-making process, improving the intelligence level of the locomotive maintenance yard.

[0053] Specifically, the intelligent analysis server uses the improved CSPDarknet as the backbone network for feature extraction, PANet for feature fusion, and uses the Anchor-based mechanism to generate prediction boxes and category probabilities in the detection head part of the YOLOV5 convolutional neural network.

[0054] When processing the locomotive sand-spreading state image, the intelligent analysis server first uses the improved CSPDarknet53 backbone network to extract image features. This step ensures the efficiency and accuracy of feature extraction. Subsequently, feature fusion is performed through PANet (Path Aggregation Network), which enhances the integration of features at different levels and improves the model's ability to recognize multi-scale and complex backgrounds. Finally, in the detection head of YOLOv5, an Anchor-based mechanism is used to generate prediction boxes and calculate category probabilities, achieving accurate positioning and classification of the locomotive sand-spreading state.

[0055] Through this advanced network structure design, the intelligent analysis server can quickly and accurately detect the sand-spreading status of the locomotive, significantly improving the reliability and practicality of the detection system. The improved CSPDarknet53 backbone network reduces the amount of calculation while maintaining the quality of feature extraction. PANet's feature fusion technology enables the model to better understand the image content and improves the recognition accuracy of the sand-spreading status of different types of locomotives. The application of the Anchor-based mechanism enables the model to show higher accuracy and robustness in target detection tasks, thereby providing strong technical support for the maintenance and operation of locomotives in the maintenance yard.

[0056] Specifically, the 180° wide-angle camera collects a number of locomotive sand-spreading images including sand-spreading and non-sand-spreading states, and at the same time, the license plate recognition device collects license plate number information and transmits it to the intelligent analysis server.

[0057] 180° wide-angle cameras are deployed on both sides of the railway tracks in the preparation yard to collect real-time images of locomotives in and out of sanding conditions. At the same time, the license plate recognition device captures and transmits the license plate number of each locomotive entering the station to the intelligent analysis server. These images and license plate information are integrated for intelligent detection and analysis of locomotive status.

[0058] By combining real-time image acquisition and license plate recognition technology, accurate monitoring and recording of the locomotive sand spreading status is achieved. This not only improves the efficiency and safety of locomotive maintenance, but also enhances the ability to identify locomotive status changes through the deep learning model of the intelligent analysis server, providing reliable data support for locomotive preventive maintenance and status assessment.

[0059] Specifically, the intelligent analysis server compares the collected license plate number information with the locomotive models corresponding to all the license plate numbers stored therein, and identifies the locomotive model information corresponding to the collected locomotive sand-spreading images in the sand-spreading and non-sand-spreading states.

[0060] The intelligent analysis server receives the license plate number information collected by the license plate recognition device and matches it with the locomotive model database stored in the server, thereby identifying the specific locomotive model corresponding to each locomotive sand spreading image. This process ensures the accurate association between image data and locomotive model information, providing the necessary model background information for subsequent image analysis and status evaluation.

[0061] Through the license plate number matching and vehicle model recognition functions of the intelligent analysis server, the system can realize personalized analysis and evaluation of the sand spreading status of locomotives of different models. This method improves the pertinence and accuracy of detection, makes maintenance work more precise, helps prevent maintenance deviations caused by vehicle model differences, and also provides data support for customized maintenance and services of locomotives, thereby improving the intelligence and automation level of maintenance yard operations.

[0062] Specifically, the intelligent analysis server determines a data set for the locomotive sand-spreading image to be input into the convolutional network neural network, and organizes it by category, wherein each category of images is stored in a corresponding folder and is divided into a training set and a validation set. The intelligent analysis server uses the albumentations library to perform random cropping, horizontal flipping, color jittering and rotation, adjusts the image to a specified input size of 224x224, and uses a bilinear interpolation algorithm to adapt the input layer of the model, normalizes the image pixel values, and scales the pixel values ​​to the range of [0,1].

[0063] The intelligent analysis server preprocesses the locomotive sand spreading images. First, the images are divided into different categories according to the different states of locomotive sand spreading, and stored in corresponding folders to form training sets and validation sets. The albumentations library is used to perform data enhancement operations such as random cropping, horizontal flipping, color jittering, and rotation on the images. Then, the image size is adjusted to the 224x224 pixels required by the convolutional neural network, and the bilinear interpolation algorithm is applied to ensure the compatibility of the image with the model input layer. Finally, the pixel values ​​are normalized and scaled to the range of [0,1] for network processing.

[0064] Through this preprocessing process of the intelligent analysis server, not only the generalization ability of the model for locomotive images in different sand spreading states is enhanced, the diversity of model training is increased through data enhancement, but also the standardization and consistency of image data are ensured, thereby improving the accuracy and robustness of the convolutional neural network in identifying the locomotive sand spreading state. In addition, the normalization process speeds up the model training speed and improves the stability of model convergence.

[0065] Specifically, the intelligent analysis server passes the preprocessed locomotive sand-spreading image as the original input to the convolutional neural network. The YOLOv5 convolutional neural network adopts CSPDarknet53 as its backbone network and performs convolution operations in its convolutional layer. The YOLOv5 convolutional neural network calculates the local features of the image by weighted addition to obtain a feature map, uses the ReLU activation function to perform nonlinear mapping on the feature map, introduces residual connections, and adopts a CSP neural network structure. The feature map is divided into two parts for alternating updates. The convolutional neural network extracts image features related to the sand-spreading state from the original image through the stacking and optimization process of the entire convolutional layer and pooling layer.

[0066] The intelligent analysis server inputs the preprocessed locomotive sand-spreading image into the YOLOv5 convolutional neural network. The network uses CSPDarknet53 as the backbone network, performs convolution operations in the convolution layer to extract local features, and generates feature maps through weighted addition. Then, the ReLU activation function is used for nonlinear mapping, and residual connections are introduced to facilitate deep network training. At the same time, the CSP structure is used to alternately update the feature map. Finally, through the multi-layer stacking of convolutional layers and pooling layers, the features closely related to the locomotive sand-spreading state are efficiently extracted from the original image.

[0067] The CSPDarknet53 backbone network and PANet feature fusion technology enhance the ability to identify the sand spreading status of the locomotive. The use of ReLU activation function and residual connection improves the network's ability to express complex images and reduces the gradient vanishing problem during deep network training. In addition, through precise feature extraction, the system can more accurately evaluate the sand spreading status of the locomotive, thereby providing reliable intelligent detection support for the maintenance yard.

[0068] Specifically, the intelligent analysis server divides a part of the locomotive sand spreading state image data set as a validation set. After each training cycle, the validation set and the test set are used to calculate the training loss and the validation loss of the YOLOv5 convolutional neural network.

[0069] If the training loss is greater than the validation loss, the intelligent analysis server performs data regularization.

[0070] During the training of the YOLOv5 convolutional neural network, the intelligent analysis server regularly divides a portion of images from the locomotive sand spreading state image dataset to form a validation set. After each training cycle, the server uses this validation set and an independent test set to evaluate the model, calculate and compare the training loss and validation loss, and monitor the performance of the model on different datasets.

[0071] When the training loss is lower than the validation loss. If this happens, the server will take data regularization measures, such as introducing L1 or L2 regularization, to reduce the overfitting of the model to the training data, thereby improving the generalization ability of the model. This ensures the stability and accuracy of the model in practical applications and improves the overall reliability of the detection system.

[0072] Specifically, when the training loss is equal to the validation loss, the intelligent analysis server runs the model on the validation set and the test set respectively, calculates the prediction results of all samples, and calculates the accuracy of the validation set and the accuracy of the test set. If the accuracy of the test set is less than the accuracy of the validation set, the number of network layers of the YOLOv5 convolutional neural network is adjusted according to the difference between the accuracy of the validation set and the accuracy of the test set.

[0073] When the training loss is equal to the validation loss, the model will be further run on the validation set and test set to predict all samples to calculate the accuracy on the two data sets. If the accuracy of the test set is found to be lower than that of the validation set, the server will analyze the accuracy difference between the two and adjust the number of network layers of the YOLOv5 convolutional neural network accordingly to optimize the model structure.

[0074] The meticulous evaluation and adjustment mechanism allows the intelligent analysis server to accurately diagnose the performance of the model on different data sets and perform targeted structural optimization. By adjusting the number of network layers, the overfitting or underfitting problems that may exist in the model can be effectively solved, and the generalization ability of the model on unknown data can be improved, thereby ensuring the accuracy and reliability of the intelligent detection system in practical applications. In addition, this method also promotes the continuous improvement of model performance and adaptability.

[0075] Specifically, the intelligent analysis server performs the same preprocessing operation on the input new locomotive sand spreading image as the training data, inputs the preprocessed image into the trained convolutional neural network, extracts the features of the input image layer by layer through the convolution layer, pooling layer and fully connected layer, generates the corresponding high-level representation, and inputs the output of the last fully connected layer into the softmax function to generate the probability output of the three-classification task, indicating the degree of locomotive sand spreading. The intelligent analysis server converts the output probability value into a classification result according to the preset range of the probability output preset internally.

[0076] If the probability value is within the preset range, it is determined that the locomotive has an excellent sand spreading degree;

[0077] If the probability value is greater than or equal to the maximum value of the preset range, it is determined that the locomotive sand spreading degree is good;

[0078] If the probability is less than the minimum value of the preset range, it is determined that the locomotive has a poor degree of sand spreading.

[0079] The intelligent analysis server performs the same preprocessing steps as the training phase on the received new locomotive sand spreading images to ensure that the image format and parameters are consistent with the training data. The preprocessed images are then fed into the trained convolutional neural network, which extracts features layer by layer through convolutional layers, pooling layers, and fully connected layers to generate a high-level representation of the image. Finally, the output of the network is converted into a probability form of a three-classification task through a softmax function, and the intelligent analysis server converts these probabilities into specific locomotive sand spreading degree classifications based on internal preset probability thresholds.

[0080] Through the classification method based on probability output and preset threshold, clear judgment criteria and operation procedures are provided for the intelligent detection of locomotive sand spreading status. It enables the system to automatically evaluate the locomotive sand spreading status and accurately classify it according to the probability output, thereby realizing real-time monitoring and evaluation of the locomotive sand spreading degree. This method not only improves the automation level of detection, but also enhances the interpretability of the system and the convenience of operation, which helps to take maintenance measures in time to ensure the safety and efficiency of railway transportation.

[0081] Specifically, the intelligent analysis server uses the Adam optimizer to adjust the model parameters to minimize the loss function and the model regularization to optimize and adjust the convolutional neural network parameters, and introduces the L1 norm penalty term through L1 regularization to sparse the parameter convolutional neural network parameters.

[0082] When training the convolutional neural network, the intelligent analysis server uses the Adam optimizer to adjust the model parameters in order to minimize the loss function and regularize the model. In this process, the server specifically introduces L1 regularization, which adds an L1 norm penalty term to the loss function to make the parameters of the convolutional neural network sparse, that is, to make the weights of some parameters approach zero.

[0083] By using the Adam optimizer combined with the L1 regularization strategy, not only the efficiency and stability of model training are improved, but also the parameter sparsification brought by L1 regularization helps to reduce model complexity, prevent overfitting, and improve the generalization ability of the model. In addition, sparse parameters may also reduce the storage requirements and computing burden of the model, making the model more suitable for deployment in resource-constrained environments. The combined application of this optimization and regularization method improves the performance and reliability of the intelligent analysis server when processing locomotive sand spreading status detection tasks.

[0084] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent detection system for sand spreading status of locomotives in a maintenance yard, characterized in that: include: Cameras are installed on both sides of the railway tracks in the maintenance yard to collect real-time images of locomotive sand spreading; License plate recognition devices are installed on both sides of the railway track at the exit of the maintenance yard, with one device on each side to collect the license plate number information of the locomotive entering the station; An LED display screen is arranged behind the license plate recognition device to prompt the locomotive to perform sand spreading work through image display; An IP audio system is installed beside the track opposite to the LED display screen to prompt the locomotive to spread sand by voice; The switch is respectively arranged in the front-end power distribution cabinet and the cabinet in the computer room, and the switch arranged in the front-end power distribution cabinet is connected to the camera, the license plate recognition device, the LED display screen and the IP audio; The switch arranged in the cabinet of the computer room is respectively connected with an optical fiber transceiver for transmitting data in the optical fiber network by converting electrical signals and optical signals, a hard disk recorder for storing historical events, a sand spreading client alarm terminal for alarming according to the sand spreading working status, and an intelligent analysis server for processing the acquired electrical signals and analyzing the sand spreading results, so as to realize the communication of all devices by building a local area network; The intelligent analysis server uses the improved CSPDarknet as the backbone network for feature extraction, PANet for feature fusion, and uses the Anchor-based mechanism to generate prediction boxes and category probabilities in the detection head part of the YOLOV5 convolutional neural network; The intelligent analysis server determines a data set according to the locomotive sand-spreading image to be input into the convolutional neural network, and organizes the data set by category, wherein each category of images is stored in a corresponding folder and is divided into two parts: a training set and a validation set. The intelligent analysis server uses the albumentations library to perform random cropping, horizontal flipping, color jittering, and rotation, adjusts the image to a specified input size of 224x224, and uses a bilinear interpolation algorithm to adapt the input layer of the model, normalizes the image pixel values, and scales the pixel values ​​to a range of [0,1] to form a preprocessed locomotive sand-spreading image; The intelligent analysis server passes the preprocessed locomotive sand-spreading image as the original input to the convolutional neural network. The YOLOv5 convolutional neural network uses CSPDarknet as its backbone network and performs convolution operations in its convolutional layer. The YOLOv5 convolutional neural network calculates the local features of the image by weighted addition to obtain a feature map, uses a ReLU activation function to perform nonlinear mapping on the feature map, introduces a residual connection, and adopts a CSP neural network structure. The feature map is divided into two parts for alternating updates. The convolutional neural network extracts image features related to the sand-spreading state from the original image through the stacking and optimization process of the entire convolutional layer and pooling layer; The intelligent analysis server divides a part of the locomotive sand spreading state image data set as a validation set. After each training cycle, the validation set and the independent test set are used to calculate the training loss and validation loss of the YOLOv5 convolutional neural network. If the training loss is greater than the validation loss, the intelligent analysis server performs data regularization; When the training loss is equal to the validation loss, the intelligent analysis server runs the model on the validation set and the independent test set respectively, calculates the prediction results of all samples, and calculates the accuracy of the validation set and the accuracy of the independent test set. If the accuracy of the independent test set is less than the accuracy of the validation set, the number of network layers of the YOLOv5 convolutional neural network is adjusted according to the difference between the accuracy of the validation set and the accuracy of the independent test set; The intelligent analysis server performs the same preprocessing operation on the input new locomotive sand spreading image as the training data, inputs the preprocessed image into the trained convolutional neural network, extracts the features of the input image layer by layer through the convolution layer, pooling layer and fully connected layer, generates the corresponding high-level representation, inputs the output of the last fully connected layer into the softmax function, generates the probability output of the three-classification task, and indicates the degree of locomotive sand spreading. The intelligent analysis server converts the output probability value into the classification result according to the preset range of the probability output preset in the server. If the probability value is within the preset range, it is determined that the locomotive has an excellent sand spreading degree; If the probability value is greater than or equal to the maximum value of the preset range, it is determined that the locomotive sand spreading degree is good; If the probability is less than the minimum value of the preset range, it is determined that the locomotive has a poor degree of sand spreading.

2. The intelligent detection system for sand spreading status of locomotives in a maintenance yard according to claim 1 is characterized in that: The camera collects a number of locomotive sand-spreading images including sand-spreading and non-sand-spreading states, and at the same time, the license plate recognition device collects license plate number information and transmits it to the intelligent analysis server.

3. The intelligent detection system for sand spreading status of locomotives in a maintenance yard according to claim 2 is characterized in that: The intelligent analysis server compares the collected license plate information with the locomotive models corresponding to all the license plate numbers stored therein, and identifies the locomotive model information corresponding to the collected locomotive sand-sprinkling images in the sand-sprinkling and non-sand-sprinkling states.

4. The intelligent detection system for sand spreading status of locomotives in a maintenance yard according to claim 3 is characterized in that: The intelligent analysis server uses the Adam optimizer to adjust the model parameters to minimize the loss function and the model regularization to optimize and adjust the parameters of the convolutional neural network, and introduces the L1 norm penalty term through L1 regularization to sparse the parameters of the convolutional neural network.

Images