A Long-Range Target Detection System Based on Multispectral Feature Fusion

Abstract

This invention discloses a long-distance track target detection system based on multispectral feature fusion, belonging to the field of rail transit inspection technology. It utilizes visible light and thermal infrared cameras to collect image data of the track area, performs dual-spectral preprocessing on the image data to improve image analysis quality, and incorporates image transmission interference risk analysis to ensure the usability of the image data. During target detection, visible light and thermal infrared feature vectors are extracted and fused. Anomalies are identified and located based on a learning model using complementary features from the fused dual-spectral data. The core of this system lies in its complementary feature space alignment fusion and anomaly model decision-making mechanism, enabling rapid identification and location of long-distance track anomalies. Furthermore, it reduces the false alarm rate through anomaly verification mechanisms across multiple distance bands, particularly for long-distance small target scenarios, achieving multiple anomaly determinations from far to near, thus improving detection reliability.

Description

Technical Field

[0001] This invention relates to the field of rail transit inspection technology, and more specifically, to a long-distance target detection system for rail transit based on multispectral feature fusion. Background Technology

[0002] In transportation systems such as high-speed trains and subways, the track is the foundation of train operation, and good track condition is the guarantee of safe train operation. Track condition and track foreign object intrusion detection technology have become one of the research hotspots in the field of rail transit automation. Most of them use cameras and image processing technology to detect scratches on the rail surface and rely on camera vision schemes to detect track foreign object intrusion.

[0003] However, due to the drastic changes in the external environment during high-speed operation of rail vehicles, such as clouds, rain, fog, haze, insufficient light at night, and sudden changes in light when rail vehicles enter or exit tunnels, the imaging quality of cameras is severely degraded, making it impossible to detect and identify the true condition of the track. As a result, it is difficult to accurately detect track intrusion targets such as falling rocks, pedestrians, and animals, which affects the safety of train operation.

[0004] To address the practical problems in existing technologies, a long-range orbital target detection system based on multispectral feature fusion is proposed. Summary of the Invention

[0005] The purpose of this invention is to solve the existing problems and provide a long-range orbital target detection system based on multispectral feature fusion compared with existing technologies.

[0006] The objective of this invention can be achieved through the following technical solution: a long-range target detection system based on multispectral feature fusion, comprising a track inspection and acquisition module, an image processing module, a transmission risk feedback module, a dual-spectral feature fusion module, a track target recognition module, an anomaly verification module, and an early warning module;

[0007] The track inspection and acquisition module is used to acquire image data of the track area in the direction of train movement. The image data includes visible light images and thermal infrared images. The image data is sent to the image processing module for dual-spectrum preprocessing.

[0008] The transmission risk feedback module is used to perform risk analysis on the transmission environment of image data, generate bispectral transmission normal signal, bispectral transmission abnormal signal and monospectral transmission abnormal signal, and send the bispectral transmission normal signal to the image processing module, and send the bispectral transmission abnormal signal and monospectral transmission abnormal signal to the early warning module.

[0009] The dual-spectral feature fusion module is used to extract features from the preprocessed visible light image and thermal infrared image, and to map the extracted visible light feature vector and thermal infrared feature vector to a unified feature space for fusion, thereby obtaining a dual-spectral feature image which is sent to the orbital target recognition module.

[0010] The orbital target recognition module uses a pre-trained and optimized anomaly feature recognition model to identify and locate anomalies in bispectral feature images, obtain anomaly region feature sets, acquire bispectral feature images in multiple distance bands, perform multi-distance anomaly feature re-identification and localization, obtain new anomaly region feature sets, and send multiple anomaly region feature sets to the anomaly verification module.

[0011] The anomaly verification module compares and verifies multiple anomaly region feature sets to determine whether the anomaly feature identification is accurate. If accurate, it generates an early warning signal and sends it to the early warning module.

[0012] Furthermore, the image processing module performs bispectral preprocessing, which includes:

[0013] For visible light images, image denoising, sharpening, and super-resolution reconstruction algorithms are used for preprocessing to enhance the contrast and clarity of visible light images;

[0014] The thermal infrared image is preprocessed with non-uniformity correction, temperature interpretation enhancement, and spatial resolution improvement to enhance its resolution.

[0015] Furthermore, the process of conducting risk analysis on the image data transmission environment includes:

[0016] The transmission risk feedback module collects image transmission interference data, including electromagnetic interference intensity, tunnel signal attenuation, vibration spectral density, and ambient temperature jump values. It substitutes the electromagnetic interference intensity, tunnel signal attenuation, and vibration spectral density values ​​into a preset visible light risk model to calculate the visible light image transmission risk value. It also substitutes the electromagnetic interference intensity, tunnel signal attenuation, vibration spectral density, and ambient temperature jump values ​​into a preset thermal infrared risk model to calculate the thermal infrared image transmission risk value. Based on the comprehensive analysis of the visible light image transmission risk value and the thermal infrared image transmission risk value, it outputs multi-signal results.

[0017] Furthermore, the process of outputting multi-signal results based on the comprehensive analysis of visible light image transmission risk value and thermal infrared image transmission risk value includes: comparing the visible light image transmission risk value and thermal infrared image transmission risk value with the corresponding preset standard risk thresholds respectively.

[0018] When both the visible light image transmission risk value and the thermal infrared image transmission risk value are greater than the corresponding preset standard risk threshold, a dual-spectral transmission abnormal signal is generated. When either the visible light image transmission risk value or the thermal infrared image transmission risk value is greater than the corresponding preset standard risk threshold, a single-spectral transmission abnormal signal is generated. When both the visible light image transmission risk value and the thermal infrared image transmission risk value are less than or equal to the corresponding preset standard risk threshold, a dual-spectral transmission normal signal is generated.

[0019] Furthermore, the feature extraction process of the orbital target recognition module for visible light images and thermal infrared images includes: extracting structural features, texture features and semantic features from visible light images, and extracting abnormal temperature features from thermal infrared images.

[0020] Furthermore, the process of training and optimizing the anomaly feature recognition model includes:

[0021] The training set is constructed by acquiring bispectral feature images under interference-free conditions, and the validation set is constructed by acquiring bispectral feature images under real interference conditions. The anomaly feature recognition model is initially trained using the training set to complete basic feature learning, and then retrained using the validation set. During the multi-stage progressive training process, the anomaly feature recognition model continuously learns different types of anomaly features to accurately identify anomaly targets in new bispectral feature images.

[0022] Furthermore, the process of comparing and verifying multiple anomaly region feature sets includes:

[0023] Multiple abnormal region feature sets are obtained, including abnormal feature vectors and abnormal feature positions. The similarity of multiple abnormal feature vectors is compared. The abnormal feature positions of multiple abnormal region feature sets obtained before and after are mapped to the same coordinate system through coordinate transformation. The abnormal feature positions are compared to see if they overlap or are within a preset deviation range threshold.

[0024] If multiple abnormal feature vectors are similar and their positions overlap or fall within a preset deviation threshold, then the abnormal feature identification is considered accurate.

[0025] This invention also proposes a method for detecting long-range targets in orbit based on multispectral feature fusion, comprising the following steps:

[0026] Step 1: Acquire visible light and thermal infrared images of the track area in the direction of train travel, and perform dual-spectrum preprocessing.

[0027] Step 2: Conduct risk analysis and early warning for the transmission environment of visible light and thermal infrared images;

[0028] Step 3: Extract features from the visible light image and the thermal infrared image, map the extracted visible light feature vector and thermal infrared feature vector to a unified feature space and fuse them to obtain a dual-spectral feature image;

[0029] Step 4: Use the pre-trained and optimized anomaly feature recognition model to identify and locate anomalies in the bispectral feature image, and obtain anomaly region feature sets in multiple range bands;

[0030] Step 5: Compare and verify the feature sets of multiple abnormal regions to determine whether the abnormal feature identification is accurate.

[0031] Compared with the prior art, the advantages of this invention are:

[0032] 1. This solution is based on the acquisition of visible light and thermal infrared images of the orbital area by visible light and thermal infrared cameras, respectively. The image data is preprocessed using dual spectra to improve the quality of image analysis. Combined with image transmission interference risk analysis, the availability of image data is ensured. During target detection, visible light and thermal infrared feature vectors are extracted and fused. Anomalies are identified and located based on a learning model that integrates complementary features from dual spectra data. Its core lies in the complementary feature space alignment fusion and anomaly model decision mechanism to achieve rapid detection, identification and location of anomalous states in long-distance orbits.

[0033] 2. Based on the above, for the abnormal feature identification process, multiple verification processes are added after the initial abnormality detection. The core idea is to reduce the false alarm rate through an abnormality verification mechanism with multiple distance zones. Especially for long-distance small target scenarios, multiple abnormality judgments from far to near are achieved to improve the reliability of track abnormality detection and to achieve accurate location of hidden abnormalities (such as internal overheating, minor structural defects, and foreign object intrusion).

[0034] 3. For the training and optimization of the anomaly feature recognition model, a training set is constructed by acquiring bispectral feature images under interference-free conditions, and a validation set is constructed by acquiring bispectral feature images under real interference conditions. The complementary features of the bispectral data are fused and the anomaly feature recognition model is initially trained using the training set to complete basic feature learning. The anomaly feature recognition model is then retrained using the validation set. During the multi-stage progressive training process, the anomaly feature recognition model continuously learns different types of anomaly features to accurately identify anomaly targets in new bispectral feature images. Attached Figure Description

[0035] Figure 1 This is a system principle block diagram of the present invention;

[0036] Figure 2 This is a flowchart of the method in Embodiment 1 of the present invention;

[0037] Figure 3 This is a flowchart of the method in Embodiment 2 of the present invention. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0039] Example 1: This invention discloses a long-range orbital target detection system based on multispectral feature fusion. Please refer to 1 and 2. Figure 2 It includes a track inspection and acquisition module, an image processing module, a transmission risk feedback module, a dual-spectral feature fusion module, a track target recognition module, an anomaly verification module, and an early warning module.

[0040] The track inspection and acquisition module is used to acquire image data of the track area in the direction of train travel. The image data includes visible light images and thermal infrared images. The image data is sent to the image processing module for dual-spectrum preprocessing, which includes:

[0041] For visible light images, image denoising, sharpening, and super-resolution reconstruction algorithms are used for preprocessing to enhance the contrast and clarity of visible light images. For thermal infrared images, non-uniformity correction, temperature interpretation enhancement, and spatial resolution enhancement are used for preprocessing to enhance the resolution of thermal infrared images.

[0042] Through a layered preprocessing mechanism, the visible light image retains track texture details even under extreme lighting conditions, and the thermal infrared image temperature resolution reaches 0.05℃, providing high-quality image details and structural information for subsequent processing, thus meeting the detection requirements of small foreign objects at long distances on high-speed trains.

[0043] In the track safety monitoring system, reliable transmission of visible light and thermal infrared images is the key to ensuring detection accuracy. The transmission risk feedback module is used to analyze the risk of the image data transmission environment and collect image transmission interference data, including electromagnetic interference intensity value, tunnel signal attenuation value, vibration spectral density value and ambient temperature jump value.

[0044] The electromagnetic interference intensity value, tunnel signal attenuation value, and vibration spectral density value are substituted into the preset visible light risk model to calculate the visible light image transmission risk value. The electromagnetic interference intensity value, tunnel signal attenuation value, vibration spectral density value, and ambient temperature jump value are substituted into the preset thermal infrared risk model to calculate the thermal infrared image transmission risk value. Based on the comprehensive analysis of the visible light image transmission risk value and the thermal infrared image transmission risk value, multi-signal results are output.

[0045] Among them, the visible light risk model is a standard model for calculating risk value by converting the interference factors affecting the transmission of visible light images into quantifiable proportional coefficients. Similarly, the thermal infrared risk model is a standard model for calculating risk value by converting the interference factors affecting the transmission of thermal infrared images into quantifiable proportional coefficients.

[0046] The process of outputting multi-signal results based on the comprehensive analysis of visible light image transmission risk values ​​and thermal infrared image transmission risk values ​​includes:

[0047] The risk values ​​for visible light image transmission and thermal infrared image transmission are compared with the corresponding preset standard risk thresholds.

[0048] When the transmission risk value of visible light image and the transmission risk value of thermal infrared image both exceed the corresponding preset standard risk threshold, a dual-spectrum transmission anomaly signal is generated. When either the transmission risk value of visible light image or the transmission risk value of thermal infrared image exceeds the corresponding preset standard risk threshold, a single-spectrum transmission anomaly signal is generated.

[0049] When the risk values ​​of visible light image transmission and thermal infrared image transmission are both less than or equal to the corresponding preset standard risk thresholds, a normal dual-spectrum transmission signal is generated. Combining the transmission characteristics of visible light and thermal infrared images, a risk assessment is conducted using a multi-dimensional risk judgment mode for transmitting interference data to ensure the availability of image data. The normal dual-spectrum transmission signal is sent to the image processing module, while the abnormal dual-spectrum transmission signal and the abnormal single-spectrum transmission signal are sent to the early warning module.

[0050] After receiving a bispectral or monospectral transmission anomaly signal, the early warning module takes countermeasures for different abnormal interference environments. For example, when strong electromagnetic interference occurs, it switches to shielded twisted-pair transmission; when tunnel signal loss occurs, it activates the vehicle edge buffer.

[0051] The dual-spectral feature fusion module responds to the normal signal of dual-spectral transmission and extracts features from visible light images and thermal infrared images. Specifically, it extracts structural features, texture features and semantic features from visible light images and abnormal temperature features from thermal infrared images.

[0052] The extracted visible light feature vector and thermal infrared feature vector are mapped to a unified feature space and fused to obtain a dual-spectral feature image, which is then sent to the orbital target recognition module.

[0053] The track target recognition module uses a pre-trained and optimized abnormal feature recognition model to identify and locate abnormal features in dual-spectral feature images (abnormal features mainly include track damage, deformation and foreign objects in the track bed), and obtains an abnormal region feature set, which includes abnormal feature vectors and abnormal feature locations. By utilizing the high spatial resolution of visible light and the physical characteristics of thermal infrared to sense the abnormal state of a track at a distance, it can quickly detect, identify and locate the abnormal state of a track.

[0054] The process of training and optimizing the anomaly feature recognition model includes:

[0055] The training set is constructed by acquiring bispectral feature images under interference-free environment, and the validation set is constructed by acquiring bispectral feature images under real interference environment. The anomaly feature recognition model is initially trained using the training set to complete basic feature learning, and the anomaly feature recognition model is retrained using the validation set. During the multi-stage progressive training process, the anomaly feature recognition model continuously learns different types of anomaly features to accurately identify anomaly targets in new bispectral feature images.

[0056] The model is initially trained using a training set (no interference data) to learn the basic features of the orbit (such as orbital structure, texture, and normal temperature distribution). The model is then fine-tuned (retrained) using a validation set (interference environment data) to improve its feature extraction and fusion capabilities under interference conditions.

[0057] At this stage, data augmentation techniques (such as simulating rain, fog, noise, and brightness changes) can be used to further expand the validation set and enhance the model's generalization ability. The focus of this stage is to optimize the model's feature extraction capabilities, ensuring that the model can accurately extract features such as edges and textures in visible light, as well as temperature distribution features in thermal infrared images. Multi-stage progressive training of the model is adopted to enhance robustness and improve the real-time detection and recognition accuracy of sudden situations such as locomotive track damage, deformation, and foreign object intrusion.

[0058] Example 2: Building upon Example 1, for the anomaly feature identification stage, multiple verification processes are added after the initial anomaly detection, as detailed below:

[0059] Please see Figure 1 and Figure 3The orbital target recognition module uses a pre-trained and optimized anomaly recognition model to identify and locate anomalies in the bispectral feature image, and obtains anomaly region feature set. This anomaly region feature set serves as the initial anomaly region feature set. The module then acquires bispectral feature images in multiple distance bands, performs multi-distance anomaly feature re-identification and re-location, and obtains a new anomaly region feature set. Finally, the module sends the multiple anomaly region feature sets to the anomaly verification module.

[0060] The anomaly verification module compares and verifies multiple anomaly region feature sets. The specific steps include: acquiring multiple anomaly region feature sets, comparing whether multiple anomaly feature vectors are similar, mapping the anomaly feature positions of multiple anomaly region feature sets obtained before and after to the same coordinate system (such as the orbital geographic coordinate system) through coordinate transformation (such as affine transformation), and comparing whether the anomaly feature positions overlap or are within a preset deviation range threshold.

[0061] If multiple abnormal feature vectors are similar and their positions overlap or fall within a preset deviation threshold, then the abnormal feature identification is considered accurate.

[0062] As can be seen from Examples 1 and 2, the present invention also proposes a method for detecting long-range targets in orbit based on multispectral feature fusion, comprising the following steps:

[0063] Step 1: Acquire visible light and thermal infrared images of the track area in the direction of train travel, and perform dual-spectrum preprocessing.

[0064] Step 2: Conduct risk analysis and early warning for the transmission environment of visible light and thermal infrared images;

[0065] Step 3: Extract features from the visible light image and the thermal infrared image, map the extracted visible light feature vector and thermal infrared feature vector to a unified feature space and fuse them to obtain a dual-spectral feature image;

[0066] Step 4: Use the pre-trained and optimized anomaly feature recognition model to identify and locate anomalies in the bispectral feature image, and obtain anomaly region feature sets in multiple range bands;

[0067] Step 5: Compare and verify the feature sets of multiple abnormal regions to determine whether the abnormal feature identification is accurate. If accurate, output the abnormal features and their locations, and generate an early warning signal to send to the early warning module. If inaccurate, start the manual review mode.

[0068] This invention involves multiple parameter thresholds. It should be noted that the thresholds, preset values, preset ranges, etc., are set for result comparison and analysis to determine whether they are good or bad. The magnitude of these thresholds is determined by a combination of large-scale model analysis of sample data and human experience. They can also be appropriately adjusted based on seasonal or common-sense influencing conditions.

[0069] In summary, the method involves acquiring image data of the orbital region using both visible light and thermal infrared cameras. Dual-spectral preprocessing is performed on the image data to improve image analysis quality. Combined with image transmission interference risk analysis, the usability of the image data is ensured. During target detection, visible light and thermal infrared feature vectors are extracted and fused. Anomalies are identified and located based on a learning model that utilizes complementary features from the fused dual-spectral data. The core of this method lies in the complementary feature space alignment and fusion, along with an anomaly model decision-making mechanism, to achieve rapid identification and location of anomalous states in distant orbits. Furthermore, an anomaly verification mechanism across multiple distance bands reduces the false alarm rate. Especially for distant, small target scenarios, multiple anomaly determinations from far to near are achieved, improving detection reliability.

[0070] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto; any equivalent substitutions or modifications made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. A long-range orbital target detection system based on multispectral feature fusion, characterized in that: It includes a track inspection and acquisition module, an image processing module, a transmission risk feedback module, a dual-spectral feature fusion module, a track target recognition module, an anomaly verification module, and an early warning module; The track inspection and acquisition module is used to acquire image data of the track area in the direction of train movement. The image data includes visible light images and thermal infrared images. The image data is sent to the image processing module for dual-spectrum preprocessing. The transmission risk feedback module is used to perform risk analysis on the transmission environment of image data, generate bispectral transmission normal signal, bispectral transmission abnormal signal and single-spectral transmission abnormal signal, and send the bispectral transmission normal signal to the image processing module. The dual-spectral feature fusion module is used to extract features from the preprocessed visible light image and thermal infrared image, and to map the extracted visible light feature vector and thermal infrared feature vector to a unified feature space for fusion, thereby obtaining a dual-spectral feature image which is sent to the orbital target recognition module. The orbital target recognition module uses a pre-trained and optimized anomaly feature recognition model to identify and locate anomalies in bispectral feature images, obtain anomaly region feature sets, acquire bispectral feature images in multiple distance bands, perform multi-distance anomaly feature re-identification and relocation from far to near, obtain new anomaly region feature sets, and send them uniformly to the anomaly verification module. The anomaly verification module compares and verifies the feature sets of multiple anomaly regions to determine whether the anomaly feature identification is accurate. If accurate, it generates an early warning signal and sends it to the early warning module. The process of conducting risk analysis on the image data transmission environment includes: The transmission risk feedback module collects image transmission interference data, including electromagnetic interference intensity, tunnel signal attenuation, vibration spectral density, and ambient temperature jump values. It substitutes the electromagnetic interference intensity, tunnel signal attenuation, and vibration spectral density values ​​into a preset visible light risk model to calculate the visible light image transmission risk value. It also substitutes the electromagnetic interference intensity, tunnel signal attenuation, vibration spectral density, and ambient temperature jump values ​​into a preset thermal infrared risk model to calculate the thermal infrared image transmission risk value. Based on the comprehensive analysis of the visible light image transmission risk value and the thermal infrared image transmission risk value, it outputs multi-signal results. The visible light risk model is a standard model for calculating risk values ​​by converting interference factors affecting visible light image transmission into quantifiable proportional coefficients. The thermal infrared risk model is a standard model for calculating risk values ​​by converting interference factors affecting thermal infrared image transmission into quantifiable proportional coefficients.

2. The orbital long-range target detection system based on multispectral feature fusion according to claim 1, characterized in that: The image processing module performs dual-spectral preprocessing, which includes the following steps: For visible light images, image denoising, sharpening, and super-resolution reconstruction algorithms are used for preprocessing to enhance the contrast and clarity of visible light images; The thermal infrared image is preprocessed with non-uniformity correction, temperature interpretation enhancement, and spatial resolution improvement to enhance its resolution.

3. The orbital long-range target detection system based on multispectral feature fusion according to claim 2, characterized in that: The process of outputting multi-signal results based on the comprehensive analysis of visible light image transmission risk value and thermal infrared image transmission risk value includes: comparing the visible light image transmission risk value and thermal infrared image transmission risk value with the corresponding preset standard risk thresholds respectively; When both the visible light image transmission risk value and the thermal infrared image transmission risk value are greater than the corresponding preset standard risk threshold, a dual-spectral transmission abnormal signal is generated. When either the visible light image transmission risk value or the thermal infrared image transmission risk value is greater than the corresponding preset standard risk threshold, a single-spectral transmission abnormal signal is generated. When both the visible light image transmission risk value and the thermal infrared image transmission risk value are less than or equal to the corresponding preset standard risk threshold, a dual-spectral transmission normal signal is generated.

4. The orbital long-range target detection system based on multispectral feature fusion according to claim 3, characterized in that: The process of feature extraction for visible light images and thermal infrared images by the orbital target recognition module includes: extracting structural features, texture features and semantic features from visible light images, and extracting abnormal temperature features from thermal infrared images.

5. The orbital long-range target detection system based on multispectral feature fusion according to claim 4, characterized in that: The process of training and optimizing anomaly feature recognition models includes: The training set is constructed by acquiring bispectral feature images under interference-free conditions, and the validation set is constructed by acquiring bispectral feature images under real interference conditions. The anomaly feature recognition model is initially trained using the training set to complete basic feature learning, and then retrained using the validation set. During the multi-stage progressive training process, the anomaly feature recognition model continuously learns different types of anomaly features to accurately identify anomaly targets in new bispectral feature images.

6. The orbital long-range target detection system based on multispectral feature fusion according to claim 1, characterized in that: The process of comparing and verifying multiple anomaly region feature sets includes: Multiple abnormal region feature sets are obtained, including abnormal feature vectors and abnormal feature positions. The similarity of multiple abnormal feature vectors is compared. The abnormal feature positions of multiple abnormal region feature sets obtained before and after are mapped to the same coordinate system through coordinate transformation. The abnormal feature positions are compared to see if they overlap or are within a preset deviation range threshold. If multiple abnormal feature vectors are similar and their positions overlap or fall within a preset deviation threshold, then the abnormal feature identification is considered accurate.

7. A method for detecting long-range orbital targets based on multispectral feature fusion, employing a long-range orbital target detection system based on multispectral feature fusion as described in any one of claims 1-6, characterized in that, Includes the following steps: Step 1: Acquire visible light and thermal infrared images of the track area in the direction of train travel, and perform dual-spectrum preprocessing. Step 2: Conduct risk analysis and early warning for the transmission environment of visible light and thermal infrared images; Step 3: Extract features from the visible light image and the thermal infrared image, map the extracted visible light feature vector and thermal infrared feature vector to a unified feature space and fuse them to obtain a dual-spectral feature image; Step 4: Use the pre-trained and optimized anomaly feature recognition model to identify and locate anomalies in the bispectral feature image, and obtain anomaly region feature sets in multiple range bands; Step 5: Compare and verify the feature sets of multiple abnormal regions to determine whether the abnormal feature identification is accurate.

Images