A method and system for detecting foreign matter in a rail train box grille
Through semantic segmentation model and image processing technology, the shape and types of foreign objects in the rail vehicle box are identified, which solves the problems of low efficiency of traditional detection methods and the inability to identify foreign objects in real time, and realizes dynamic monitoring of foreign objects in the rail train box grille and Classification improves safety and detection efficiency.
Patent Information
- Application Number
- CN202411148370.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-21
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-08-21
AI Technical Summary
Traditional rail vehicle detection methods are inefficient and cannot identify foreign species in real time, making it difficult to meet the monitoring needs of foreign objects for vehicle safety during high-speed driving.
Semantic segmentation model and image processing technology are used to analyze the morphology of foreign objects in the cage by obtaining the bottom image of the vehicle, and determine that the foreign object is adsorbed or suspended, and index the foreign object species in the database according to the deformation ability or swing amplitude to achieve dynamic monitoring and real-time identification.
Real-time monitoring and classification of foreign objects in the rail train box grille is realized, detection efficiency is improved, potential threats can be known in a timely manner in the backend, and manual inspection workload and maintenance costs are reduced.
Smart Images

Figure CN118658016B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of rail vehicle box grille detection, and more particularly to a method and system for detecting foreign matter in a rail vehicle box grille. Background Art
[0002] The safety of rail vehicle operation has always attracted widespread attention. However, traditional rail vehicle detection methods mainly rely on manual inspections, which have problems such as low efficiency and susceptibility to human interference. Therefore, rail vehicle anomaly detection technology based on machine vision is becoming a hot topic in research and application fields.
[0003] Generally, there are thousands of parts installed on the bottom of rail vehicles. Many trains need to be inspected and maintained every day. The safety status inspection of vehicle parts is an extremely arduous task. Currently, it includes manual inspection and machine vision inspection. Compared with manual inspection, machine vision inspection has better accuracy and efficiency. The existing machine vision inspection includes two parts. One part requires the train to drive the vehicle to the inspection area and park it. The inspection robot inspects it in sequence along the set track. The other part requires the train to drive through the gantry for 360-degree inspection. Using high-resolution images or video data, with the help of image processing and semantic segmentation, the target in the image is compared with the standard in the database for similarity to determine whether there are abnormal objects in the image. However, both of the above methods require the train to be in a certain specific form for inspection. When the train is running at high speed, if foreign objects are adsorbed in the cage for a long time, it is easy to cause the fresh air system or cage in the train to be damaged. In addition, the existing detection cannot summarize the types of foreign objects in real time, making it impossible for the background to accurately know the type of foreign objects, which is difficult to meet the existing detection methods. Summary of the invention
[0004] In view of the deficiencies in the prior art, an object of the present invention is to provide a method and system for detecting foreign objects in a rail train box grille.
[0005] To achieve the above object, the present invention provides the following technical solutions:
[0006] A method for detecting foreign matter in a rail train box grille comprises the following steps:
[0007] A train bottom image acquisition step, acquiring a visually captured train bottom image as an image to be analyzed;
[0008] The step of analyzing the morphology of foreign matter in the cage is to determine whether there is foreign matter in the cage area by using a semantic segmentation model on the image to be analyzed. If so, continuous position analysis is performed on the foreign matter to obtain the morphology of the foreign matter, which includes an adsorbed morphology and a suspended morphology.
[0009] an adsorbed foreign matter type analysis step, when the foreign matter is in an adsorbed form, obtaining an image to be analyzed when the net box fan reduces the suction intensity, obtaining a deformation capacity value of the foreign matter in the image to be analyzed by position analysis, and indexing the foreign matter type in a database based on the deformation capacity value;
[0010] The step of analyzing the type of swinging foreign objects is as follows: when the foreign object is in a suspended form, images to be analyzed of the train at different speeds are obtained, the swing amplitude of the foreign object is obtained by dynamic contour analysis in the images to be analyzed, and the type of foreign object is indexed in the database based on the swing amplitude.
[0011] Furthermore, the cage foreign body morphology analysis step includes a foreign body position marking sub-step.
[0012] In the foreign body position marking sub-step, the image to be analyzed is cut through a preset semantic segmentation model to obtain a cage area map, the cage area map is processed through a binary image to obtain a binarized image, the foreign body area is obtained based on pixel judgment in the binarized image, and the outer contour line of the foreign body area is extracted through a contour detection algorithm, the degree of fit of the outer contour line with the cage background in the binarized image is calculated, and the foreign body is judged as being in an adsorbed form or a suspended form based on the fit.
[0013] Furthermore, the cage foreign body morphology analysis step also includes a foreign body area coverage analysis sub-step.
[0014] The foreign body area coverage analysis sub-step maps the outer contour line in the image to be analyzed, obtains various extended contour lines through edge detection of the connecting parts of the outer contour line in the image to be analyzed, selects the extended contour line that connects with the outer contour line from each extended contour line, and splices it with the outer contour line to obtain the overall contour line of the foreign body, calculates the coverage area of the foreign body on the grid and the hanging area of the foreign body beyond the grid according to the outer contour line and the extended contour line, then compares the coverage area and the hanging area to obtain the area ratio, and judges whether the foreign body is in an adsorbed form or a hanging form according to the area ratio.
[0015] Furthermore, in the step of analyzing the types of adsorbed foreign matter, the image to be analyzed is obtained when the mesh box fan has reduced suction intensity, and it is analyzed whether the outline of the foreign matter can be extracted. If so, the degree of fit of the foreign matter at the normal suction intensity of the mesh box fan is compared with the degree of fit of the foreign matter at the reduced suction intensity of the mesh box fan as the deformation capacity value. If not, the foreign matter type analysis is ignored.
[0016] Furthermore, in the step of analyzing the type of adsorbed foreign matter, the area, number and position of the foreign matter area on the mesh box are determined based on the outer contour lines of the foreign matter. If the number of foreign matter areas is greater than 2, and the foreign matter area is located at the mesh of the mesh box and the area is the same as the mesh area of the mesh box, then the type of foreign matter can be indexed in the database as mud stains.
[0017] Furthermore, in the step of analyzing the type of swinging foreign matter, images to be analyzed are obtained when the train is traveling at a high speed and images to be analyzed when the train is traveling at a low speed, and the hanging area of the extended contour line and the relative position of the extended contour line and the cage are analyzed as the swing amplitude in the high-speed and low-speed states respectively.
[0018] Furthermore, in the swinging foreign object type analysis step, an extended contour line of the train when traveling at a low speed is obtained, and whether it contacts the track is determined based on the extended contour line. If so, a crisis instruction is output; if not, the type of foreign object is indexed in a database based on the swing amplitude.
[0019] Furthermore, the step of analyzing the types of adsorbed foreign matter also includes an auxiliary analysis sub-step, and the auxiliary analysis sub-step obtains the reflectivity of the light source irradiating the mesh box or obtains the transmittance of the light source irradiating the mesh box, and indexes the types of foreign matter in the database according to the reflectivity or transmittance.
[0020] Furthermore, it also includes a foreign object monitoring feedback step, in which the foreign object detection feedback step obtains the air intake volume of the grid box fan of each train car, selects the image to be analyzed of the train car with a larger difference in the air intake volume, and analyzes whether there are foreign objects in the image to be analyzed of the train car. If so, accurate instructions are fed back, otherwise, wrong instructions are fed back.
[0021] A foreign body detection system for a rail train box grille, comprising:
[0022] The database construction module classifies foreign objects according to their weight, deformation ability and viscosity, and builds a database of different foreign object categories;
[0023] The wind press image acquisition module acquires the visually collected image as the image to be analyzed;
[0024] The cage foreign body morphology analysis module uses the semantic segmentation model to determine whether there are foreign bodies in the cage area of the image to be analyzed. If so, continuous position analysis is performed on the foreign bodies to obtain the foreign body morphology, which includes adsorption morphology and hanging morphology.
[0025] An adsorbed foreign matter type analysis module, when the foreign matter is in an adsorbed form, obtains an image to be analyzed when the net box fan reduces the suction intensity, obtains the deformation capacity value of the foreign matter in the image to be analyzed by position analysis, and indexes the foreign matter type in the database based on the deformation capacity value;
[0026] The swinging foreign body type analysis module obtains the train images to be analyzed at different speeds when the foreign body is in a suspended form, obtains the swing amplitude of the foreign body in the image to be analyzed through contour dynamic analysis, and indexes the foreign body type in the database based on the swing amplitude.
[0027] The beneficial effects of the present invention are as follows: by judging the shape of the foreign object at its location, and then controlling the fan intensity according to the corresponding foreign object shape or judging the swing amplitude of the foreign object at different vehicle speeds to further judge the type of the foreign object, compared with traditional detection area detection or 360 detection, firstly, it can realize dynamic monitoring of the train at any time period to eliminate foreign objects with a higher degree of danger, and secondly, it can further judge the classification type of the foreign object to ensure that the background can know in real time whether the foreign object will cause a threat. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 It is the overall flow chart of the present invention;
[0029] Figure 2 It is a logic diagram for judging adsorbed foreign matter or suspended foreign matter in the present invention;
[0030] Figure 3 is a binary image of adsorbed foreign matter in the present invention;
[0031] Figure 4 It is a display diagram of the sticky foreign matter in the present invention. DETAILED DESCRIPTION
[0032] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. The same parts are represented by the same reference numerals. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to directions in the accompanying drawings, and the words "bottom surface" and "top surface", "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.
[0033] At present, foreign body detection at the train air compressor can only be performed in a certain form, such as driving to the detection area or passing through the 360-degree detection gantry. When the train is running at high speed, if foreign objects are adsorbed in the mesh box for a long time, it is easy to cause the fresh air system or mesh box in the train to be damaged. In addition, the existing detection cannot summarize the types of foreign objects in real time, so that the background cannot accurately know the type of foreign objects, which is difficult to meet the existing detection methods. Therefore, the present invention designs a foreign body detection method for the box grille of a rail train, such as Figure 1 As shown, the following steps are included:
[0034] The train bottom image acquisition step includes installing a high-definition camera at an appropriate position of the train (such as the bottom or side) to ensure that the wind press area, especially the cage area, including the area around the cage area, is covered, and the camera is started to capture the image of the walking part regularly or in real time as the image to be analyzed, which also includes an image preprocessing process to remove noise and enhance contrast of the collected image to be analyzed so as to achieve optimal resolution and brightness;
[0035] The steps of analyzing the morphology of foreign bodies in the cages are as follows: before this, a large amount of historical data of the cage grilles under different environments is obtained, and the cage model is trained, tested and verified according to the historical data. The pre-processed image is scanned using a pre-trained semantic segmentation model (such as YOLO, FasterR-CNN, etc.), and the position of the cage is identified from the image to be analyzed. Then, it is cropped to highlight the cage area and reduce interference factors. Then, the area of the cut cage position is used to determine whether there are foreign bodies. If so, the position of the detected foreign bodies is continuously tracked, and the movement characteristics of the foreign bodies are determined by judging the current position of the foreign bodies relative to the cages, or by comparing the position information at different time points, so as to determine their morphology. The morphology of foreign bodies includes adsorption morphology and hanging morphology.
[0036] The step of analyzing the type of adsorbed foreign matter is that when the foreign matter is in the adsorbed form, the control system adjusts the suction intensity of the cage fan to a lower level. After reducing the suction intensity, the image of the cage area is collected again. By comparing the position or shape changes of the foreign matter before and after the adjustment, its deformation capacity value is calculated. The deformation capacity value reflects the degree of response of the foreign matter to the change of wind pressure. Then, the deformation capacity value is used as a query condition to index the corresponding foreign matter type in the database. The database should pre-store the deformation characteristics of different types of foreign matter under different wind pressure conditions. For example, the database includes sheet-like soft foreign matter (paper type), sticky foreign matter, etc.
[0037] The steps for analyzing the type of swinging foreign objects are as follows: when the foreign object is in a suspended form, the images to be analyzed of the train at different speeds are obtained. The contours of the foreign objects at different speeds are extracted from the images to be analyzed through image processing technology, and the swing amplitude is calculated. The swing amplitude is an important indicator for evaluating the quality, shape and suspension method of foreign objects. The swing amplitude is used as a query condition to search for the corresponding foreign object type in the database, which contains animal carcasses, domestic garbage, etc.
[0038] Finally, a report is generated with information such as the type, location, and shape of the detected foreign object and sent to the train control system, which is also copied to the maintenance personnel, who then formulate corresponding countermeasures based on the type and nature of the foreign object.
[0039] like Figure 2 and Figure 3 As shown, the mesh cage foreign body morphology analysis step includes a foreign body position marking sub-step, specifically, including the first step, mesh cage area cutting: using a preset semantic segmentation model (deep learning model) to scan the image to be analyzed, identify and cut out the mesh cage area, and generate a mesh cage area map;
[0040] The second step is binary image processing: Binarization is performed on the cage area map, and the image is converted into an image containing only black and white colors, that is, a binary image is generated, in which black represents foreign matter and the mesh of the cage, and white represents the cage grille;
[0041] The third step is foreign body area identification: in the binary image, the foreign body area is determined and marked according to the pixel value. Generally, a set of adjacent black pixels is regarded as a foreign body area, but it does not exclude the possibility of multiple foreign bodies in the subsequent specific analysis of this situation.
[0042] The fourth step is contour detection and extraction: use the contour detection algorithm (contour search algorithm after Canny edge detection) to find edge pixels, connect the found edge pixels to form a continuous boundary line as the contour line of the foreign body area, which also includes contour optimization or removal of false contours.
[0043] The fifth step is to calculate the degree of fit: the fit between the extracted contour line and the cage background is calculated. The fit can be based on the proximity between the contour line and the cage edge or the cage background image as the reference plane to analyze whether there is an unclear intersection or distance between the contour line and the cage background image. Since the camera provides light when shooting, when the foreign object is completely attached to the cage, the contour line of the foreign object is relatively clear, and it can be judged from the visual shooting angle that the foreign object is completely attached to the cage. When the foreign object is not partially attached to the cage, the light source may shine into the space between the foreign object and the cage, and it can be judged based on the pixels. If it fails to shine, it is necessary to combine the visual shooting angle to judge whether the foreign object is not attached.
[0044] Step 6: Shape judgment: According to the preset threshold or judgment logic, the fit value is used to judge the shape of the foreign body. If the fit is higher than a certain threshold, it is judged as an adsorption shape, otherwise, it is judged as a hanging shape.
[0045] The above-mentioned foreign body position marking sub-step is mainly to detect the shape of foreign bodies in the cage area, and ignore the foreign body parts outside the cage area. Of course, if there are foreign bodies outside the cage area, but the exceeding area is not large, it can be judged as an adsorbed foreign body. If the exceeding area is too large and it swings, it needs to be judged as a hanging foreign body, which has the risk of falling at any time. Therefore, if Figure 2 As shown, in order to detect foreign bodies that exceed the cage part, the cage foreign body morphology analysis step also includes a foreign body area coverage analysis sub-step, specifically, including the first step, shape contour mapping: accurately mapping the obtained shape contour to the image to be analyzed, and determining the position of the foreign body in the image to be analyzed;
[0046] The second step is contour extension: in the image to be analyzed, edge detection technology is used to find the part that connects with the outer contour line, and various extended contour lines are generated. These extended contour lines may represent the hanging part of the foreign body due to exceeding the net box. The extended contour line may contain the contour lines of other parts. Therefore, it is necessary to select the extended contour line that connects with the outer contour line from the various extended contour lines, and splice it with the outer contour line to obtain the overall contour line of the foreign body. The selection is based on the fact that the outer contour line and the extended contour line will contain an edge line of the net box at the same time;
[0047] Step 3: Area calculation: Calculate the area covered by the foreign matter on the grid (covered area) based on the projection of the outline on the cage, and calculate the area of the part where the outline extends beyond the cage (hanging area);
[0048] The fourth step is area ratio and morphology judgment: By comparing the ratio of the covered area to the hanging area (area ratio), the morphology of the foreign matter can be judged. For example, if the hanging area accounts for a large proportion, it may mean that the foreign matter mainly exists in a hanging form; conversely, if the covered area accounts for a large proportion, the foreign matter may be mainly adsorbed on the mesh cage.
[0049] The beneficial effects of the above-mentioned foreign body morphology analysis include: 1. By accurately identifying and judging the morphology of foreign bodies in the cage, potential safety hazards can be discovered in time. For foreign bodies in adsorbed form, it may mean that they have caused some degree of damage or impact on the train or cage; and for foreign bodies in suspended form, they may fall off as the train runs, causing harm to the train or the surrounding environment. Therefore, this analysis helps to take measures in advance to avoid accidents; 2. Automated foreign body detection and morphology analysis can greatly reduce the workload of manual inspections, improve maintenance efficiency, and can quickly locate and handle foreign bodies, reducing downtime and maintenance costs; 3. Through detailed data analysis and processing, the system can provide comprehensive decision-making support for maintenance personnel. For example, based on the type, morphology and location of the foreign body, the system can recommend the most appropriate treatment plan, including cleaning methods, maintenance measures, etc., which helps maintenance personnel make more accurate and timely decisions.
[0050] like Figure 1 As shown, the steps for analyzing the types of adsorbed foreign matter are to obtain images of the net box fan at normal suction intensity and reduced suction intensity, perform necessary preprocessing on the images, clearly identify and analyze the boundaries between the foreign matter and the background, and then use the contour detection method to detect whether the contour line of the foreign matter can be extracted from the image to be analyzed. If it can be extracted, calculate the foreign matter fit under normal suction intensity and the foreign matter fit under reduced suction intensity respectively, compare the two calculated fits to determine the deformation capacity value, and evaluate the deformation capacity of the foreign matter under different suction intensities according to the size of the deformation capacity value. A larger deformation capacity value may mean that the foreign matter is more susceptible to changes in suction intensity, while a smaller value indicates that the foreign matter is more stable. For example, when the foreign matter is a thin plastic bag, it is easily affected by suction intensity, and when the foreign matter is sticky garbage, it is relatively less affected by suction intensity than a thin plastic bag. When it is a thin plastic bag, you can consider turning off the net box fan to make the garbage bag fall off automatically. When it is sticky garbage, feedback is required for subsequent treatment measures.
[0051] In addition, if the outline of the foreign matter cannot be extracted, it can be explained that the foreign matter fell off due to the reduction of suction intensity, and the analysis of the type of foreign matter can be ignored.
[0052] The step of analyzing the type of adsorbed foreign matter is to calculate the area of the enclosed area of each screened contour, count the total number of foreign matter areas in the cage area image, determine the specific position of each foreign matter area on the cage, especially the relative position relationship with the mesh hole. According to historical data feedback, when the mesh hole is blocked by mud stains on the cage, the number of blocked mesh holes is usually large. Therefore, when the number of foreign matter areas is greater than 2 and these areas are all located at the mesh holes of the cage, the next step is to compare the area of the foreign matter area with the area of the cage mesh hole. If the areas of the two are similar (a certain tolerance range can be set), it is preliminarily judged that the foreign matter may be closely attached to the mesh hole. According to the above analysis results, the corresponding foreign matter type is indexed in the pre-established database. If specific conditions (such as area, quantity, location, etc.) are met and there are corresponding records in the database, the foreign matter type is indexed as "mud stains".
[0053] In addition, it is not ruled out that there is a large amount of mud on the surface of the mesh box. In this case, the corresponding foreign matter should be adsorbed foreign matter. The foreign matter can be determined as "mud stains" by judging its color, whether it is affected by the suction intensity, whether it is affected by the vehicle speed, and whether there is a height difference between the foreign matter layer at the mesh and the foreign matter layer between the meshes.
[0054] The step of analyzing the type of adsorbed foreign matter also includes an auxiliary analysis sub-step. Specifically, the type of foreign matter can be identified by its transmittance or reflectance. For example, films, clothing or metal objects have different transmittances and reflectances. Therefore, it is necessary to establish a database containing different types of foreign matter and their corresponding reflectance / transmittance ranges, and compare the measured reflectance / transmittance data with the data in the database. Based on the comparison results, the most matching type of foreign matter is indexed in the database, and the results of foreign matter contour extraction are combined with the results of reflectance / transmittance analysis for comprehensive analysis. This helps to more accurately determine the type of foreign matter, especially when the contours are similar but the materials are different.
[0055] like Figure 1As shown, the steps for analyzing the types of swinging foreign objects are to obtain images to be analyzed when the train is traveling at high speed and images to be analyzed when the train is traveling at low speed, and respectively analyze the hanging area of the extended contour line and the relative position of the extended contour line and the mesh box in the high-speed traveling state and the low-speed traveling state as the swing amplitude. The weight of the foreign object can be judged by the swing amplitude. The heavier the foreign object is, the smaller the swing amplitude of the hanging part is when the speed increases, and the foreign object with relatively light weight has a smaller swing amplitude. In addition, combined with other features that may be extracted from the image (such as the shape, size, texture, etc. of the foreign object), possible types of foreign objects are indexed in a pre-established database. The database should contain information such as the swing amplitude range and feature description of different types of foreign objects. A matching algorithm (such as nearest neighbor search, fuzzy matching, etc.) is used to find the entry that best matches the current foreign object feature in the database, and the type of foreign object under the entry is output to the background as the detected foreign object type.
[0056] The step of analyzing the type of swinging foreign objects is to obtain the extended contour line of the train when it is running at a low speed, and determine whether it touches the track based on the extended contour line. If so, a crisis instruction is output. If not, the type of foreign object is indexed in the database according to the swing amplitude, wherein the extended contour line includes the maximum range that the foreign object may reach during the swinging process. Specifically, the extracted extended contour line is compared with the boundary of the track. The boundary of the track needs to be determined, and then the minimum distance between the contour line and the boundary of the track is calculated. If the minimum distance is less than the preset safety threshold, it is considered that the foreign object may touch the track, and feedback needs to be given to the background. Based on the above database search, the indexed foreign object type is output, and according to the characteristics and potential risks of the foreign object type, corresponding response measures are formulated or recommended, including continuous monitoring, increasing the inspection frequency, taking protective measures or notifying relevant departments to handle it.
[0057] like Figure 1 As shown, it also includes a foreign matter monitoring feedback step. Due to different scenarios, the air intake volume may deviate from the standard preset air intake volume. Therefore, the air intake volume of each carriage is monitored. Specifically, the air intake volume of the box fan of each train carriage detected by the flow sensor is obtained, and the image to be analyzed of the train carriage with a large difference in air intake volume is selected, and the image to be analyzed of the train carriage is analyzed to see whether there is foreign matter. If so, an accurate instruction is fed back, otherwise, an error instruction is fed back.
[0058] A foreign body detection system for a rail train box grille, comprising:
[0059] The database construction module classifies foreign objects according to their weight, deformation ability and viscosity, and builds a database of different foreign object categories;
[0060] The wind press image acquisition module acquires the visually collected image as the image to be analyzed;
[0061] The cage foreign body morphology analysis module uses the semantic segmentation model to determine whether there are foreign bodies in the cage area. If so, the foreign bodies are continuously analyzed for their positions to obtain their morphology, which includes adsorption morphology and hanging morphology.
[0062] The adsorbed foreign matter type analysis module, when the foreign matter is in the adsorbed form, obtains the image to be analyzed when the net box fan reduces the suction intensity, obtains the deformation capacity value of the foreign matter in the image to be analyzed by position analysis, and indexes the foreign matter type in the database based on the deformation capacity value;
[0063] The swinging foreign body type analysis module obtains the train images to be analyzed at different speeds when the foreign body is in a suspended form. The swing amplitude of the foreign body is obtained through contour dynamic analysis in the image to be analyzed, and the type of foreign body is indexed in the database based on the swing amplitude.
[0064] The above are only preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.
Claims
1. A method for detecting foreign matter in a rail train box grille, comprising the following steps: A train bottom image acquisition step, acquiring a visually captured train bottom image as an image to be analyzed; The step of analyzing the morphology of foreign matter in the cage is to determine whether there is foreign matter in the cage area by using a semantic segmentation model on the image to be analyzed. If so, continuous position analysis is performed on the foreign matter to obtain the morphology of the foreign matter, which includes an adsorbed morphology and a suspended morphology. The step of analyzing the type of adsorbed foreign matter is as follows: when the foreign matter is in an adsorbed form, an image to be analyzed is obtained when the net box fan reduces the suction intensity, a deformation capacity value of the foreign matter is obtained by position analysis in the image to be analyzed, and the type of foreign matter is indexed in the database based on the deformation capacity value; an image to be analyzed when the net box fan reduces the suction intensity is obtained, and an analysis is performed to determine whether the contour line of the foreign matter can be extracted; if so, a comparison is made between the degree of fit of the foreign matter under the normal suction intensity of the net box fan and the degree of fit of the foreign matter under the reduced suction intensity of the net box fan as the deformation capacity value; if not, the analysis of the type of foreign matter is ignored; The step of analyzing the type of swinging foreign objects is as follows: when the foreign object is in a suspended form, images to be analyzed of the train at different speeds are obtained, the swing amplitude of the foreign object is obtained by dynamic contour analysis in the images to be analyzed, and the type of foreign object is indexed in the database based on the swing amplitude.
2. According to claim 1, a method for detecting foreign matter in a rail train box grille is characterized in that: The cage foreign body morphology analysis step includes a foreign body position marking sub-step, In the foreign body position marking sub-step, the image to be analyzed is cut through a preset semantic segmentation model to obtain a cage area map, the cage area map is processed through a binary image to obtain a binarized image, the foreign body area is obtained based on pixel judgment in the binarized image, and the outer contour line of the foreign body area is extracted through a contour detection algorithm, the degree of fit of the outer contour line with the cage background in the binarized image is calculated, and the foreign body is judged as being in an adsorbed form or a suspended form based on the fit.
3. According to claim 2, a method for detecting foreign matter in a rail train box grille is characterized in that: The cage foreign body morphology analysis step also includes a foreign body area coverage analysis sub-step. The foreign body area coverage analysis sub-step maps the outer contour line in the image to be analyzed, obtains various extended contour lines by edge detection of the connecting parts of the outer contour line in the image to be analyzed, selects the extended contour line that connects with the outer contour line from each extended contour line, and splices it with the outer contour line to obtain the overall contour line of the foreign body, calculates the coverage area of the foreign body on the grid and the hanging area of the foreign body beyond the grid according to the outer contour line and the extended contour line, then compares the coverage area and the hanging area to obtain the area ratio, and judges whether the foreign body is in an adsorbed form or a hanging form according to the area ratio.
4. The method for detecting foreign matter in a rail vehicle box grille according to claim 1, characterized in that: The step of analyzing the type of adsorbed foreign matter determines the area, number and position of the foreign matter area on the mesh box based on the outer contour of the foreign matter. If the number of foreign matter areas is greater than 2, the foreign matter area is located at the mesh of the mesh box and the area is the same as the mesh area of the mesh box, then the type of foreign matter can be indexed in the database as mud stains.
5. The method for detecting foreign matter in a rail vehicle box grille according to claim 3, characterized in that: The step of analyzing the type of swinging foreign matter obtains the images to be analyzed when the train is in a high-speed driving state and the images to be analyzed when the train is in a low-speed driving state, and analyzes the hanging area of the extended contour line in the high-speed driving state and the low-speed driving state and the relative position of the extended contour line and the net box as the swing amplitude.
6. The method for detecting foreign matter in a rail vehicle box grille according to claim 5, characterized in that: The swinging foreign body type analysis step obtains the extended contour line of the train when it is running at a low speed, and determines whether it touches the track based on the extended contour line. If so, a crisis instruction is output; if not, the foreign body type is indexed in the database based on the swing amplitude.
7. The method for detecting foreign matter in a rail vehicle box grille according to claim 4, characterized in that: The step of analyzing the type of adsorbed foreign matter also includes an auxiliary analysis sub-step, in which the reflectivity of the light source irradiating the mesh box or the transmittance of the light source irradiating the mesh box is obtained, and the type of foreign matter is indexed in the database according to the reflectivity or transmittance.
8. The method for detecting foreign matter in a rail train box grille according to claim 1, characterized in that: The method further includes a foreign object monitoring and feedback step, wherein the foreign object detection and feedback step obtains the air intake volume of the grid box fan of each train compartment, selects the image to be analyzed of the train compartment with a larger difference in the air intake volume, and analyzes whether there is a foreign object in the image to be analyzed of the train compartment. If so, an accurate instruction is fed back, and if not, an error instruction is fed back.
9. A foreign body detection system for a rail train box grille, characterized in that: include The database construction module classifies foreign objects according to their weight, deformation ability and viscosity, and builds a database of different foreign object categories; The wind press image acquisition module acquires the visually collected image as the image to be analyzed; The cage foreign body morphology analysis module uses the semantic segmentation model to determine whether there are foreign bodies in the cage area of the image to be analyzed. If so, continuous position analysis is performed on the foreign bodies to obtain the foreign body morphology, which includes adsorption morphology and hanging morphology. The adsorbed foreign matter type analysis module, when the foreign matter form is an adsorbed form, obtains the image to be analyzed under the mesh box fan with reduced suction intensity, obtains the deformation capacity value of the foreign matter in the image to be analyzed by position analysis, and indexes the foreign matter type in the database based on the deformation capacity value; obtains the image to be analyzed under the mesh box fan with reduced suction intensity, analyzes whether the contour line of the foreign matter can be extracted, and if so, compares the degree of fit of the foreign matter under the mesh box fan's normal suction intensity with the degree of fit of the mesh box fan under reduced suction intensity as the deformation capacity value, if not, ignores the foreign matter type analysis; The swinging foreign body type analysis module obtains the train images to be analyzed at different speeds when the foreign body is in a suspended form, obtains the swing amplitude of the foreign body in the image to be analyzed through contour dynamic analysis, and indexes the foreign body type in the database based on the swing amplitude.
Citation Information
Patent Citations
A small sample terahertz image foreign matter detection method based on integrated deep learning
CN109948527A
Motor train unit train side foreign matter detection method
CN112288717A
Laundry treating apparatus
CN114222838A
Railway vehicle anomaly detection method and system
CN117557570A
Foreign matter detection method, device and system
CN117994542A
