Airborne earth surface mine and unexploded ammunition detection system and method based on image recognition

Abstract

The invention discloses an airborne earth surface mine and unexploded ammunition detection system and method based on image recognition, and particularly relates to the technical field of image recognition. The method comprises the following steps: monitoring wireless signal intensity in real time, analyzing signal fluctuation and image transmission delay, evaluating the influence of delay on image resolution and detail maintenance, further analyzing the compression quality of an image, dividing the image into a slight decline type and a serious decline type according to the decline degree of the image quality, and taking corresponding treatment measures; the image quality is improved through local enhancement; for serious decline, the accuracy of an image recognition algorithm is predicted and early warned, the algorithm is automatically adjusted to improve the detection accuracy, the accuracy of earth surface thunder and unexploded ammunition detection can be effectively improved, and it is ensured that real-time and stable image recognition can still be achieved in a complex environment.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to an airborne surface mine and unexploded ordnance detection system and method based on image recognition. Background Art

[0002] Surface mines and unexploded ordnance (UXO) are a common safety hazard in military training and infrastructure construction. Traditional methods for removing surface mines and UXO rely on manual approach and detection, which not only poses significant safety risks but also limits the efficiency and accuracy of removal operations. With the development of drone technology, combined with remote control and aerial image acquisition capabilities, drones have become a valuable tool for removing and destroying surface mines and UXO. Drones can perform global or local image acquisition in hazardous areas and use advanced image recognition technology to identify the type and status of surface mines and UXO in real time, providing a precise basis for removal and destruction plans. Furthermore, drones can perform efficient risk assessment and surveillance without human intervention, significantly improving the safety of EOD personnel.

[0003] The introduction of image recognition technology has further enhanced the intelligence of drone-based bomb disposal systems. By transmitting high-resolution images and comparing them with databases, the system can quickly identify and classify different types of surface mines and unexploded ordnance, determine their hazard level, and assess their disposal difficulty. This drone-based detection system, powered by image recognition, not only reduces manual intervention but also accurately obtains target information in the shortest possible time. When performing disposal tasks, drones can use real-time image data to dispatch appropriate shaped charges for detonation and destruction, effectively preventing direct contact between personnel and surface mines and unexploded ordnance, and providing precise and safe technical support for their disposal.

[0004] The existing technology has the following shortcomings:

[0005] Wireless transmission of high-resolution images consumes significant bandwidth, especially over long distances, and real-time, stable data transmission may not be guaranteed. Because drones may be at high altitudes, the signal between ground-based receiving equipment and the drone may be affected by terrain, weather, and other wireless signal interference. Transmission can result in delays, data loss, or compression, leading to reduced image quality. Furthermore, reduced image quality directly impacts the accuracy of image recognition algorithms, potentially misidentifying surface mines and unexploded ordnance as other objects or even failing to identify them. Summary of the Invention

[0006] The purpose of the present invention is to provide an airborne surface mine and unexploded ordnance detection system and method based on image recognition to address the shortcomings of the background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an airborne surface mine and unexploded ordnance detection method based on image recognition, comprising the following steps:

[0008] S1: The drone flies over the target area through a remote control command system. The drone uses its onboard high-resolution camera to collect real-time images of surface mines, unexploded ordnance, and their surroundings. It also measures and records the wireless communication signal strength between the drone and the ground receiver.

[0009] S2: Compare and analyze signal strength fluctuations and image transmission delays to determine the extent to which delays interfere with image resolution and detail preservation, and assess the impact of delays on image real-time performance.

[0010] S3: If the delay is severe, the stored compression ratio information of each frame of the image during the image compression process is analyzed, and the difference between the compressed image and the original image is compared to evaluate the image compression quality;

[0011] S4: Evaluate the degree of image quality degradation based on the impact of the delay on image real-time performance and image compression quality, and classify the image quality degradation into severe degradation and slight degradation based on the evaluation results;

[0012] S5: For minor drops, the image quality is improved by locally enhancing the image, making the features of surface mines and unexploded ordnance clearer. For severe drops, the accuracy of the image recognition algorithm within a fixed time period is predicted and warned, and the image recognition algorithm is automatically adjusted based on the prediction results to improve the accuracy of surface mine and unexploded ordnance detection.

[0013] Preferably, in S2, a detail loss anomaly index is generated by comparing the degree of detail loss of the compressed image and the original image due to image transmission delay. The method for obtaining the detail loss anomaly index is:

[0014] Get the original image and the compressed image, match the size of the two images, and perform grayscale processing on the two images. The structural similarity index SSIM is used to measure the similarity between the two images, taking into account three aspects: brightness, contrast, and structure. The formula is: Where x and y are the corresponding pixel values ​​of the original image and the compressed image, μ x and μ y is the local mean of the original image and the compressed image, that is, the average value of the image block, and is the local variance of the original image and the compressed image, indicating the degree of brightness change in the local area, σ xyis the local covariance between the original image and the compressed image, indicating the degree of similarity between the two images; C1 and C2 are constants used to avoid the denominator being zero, where L is the dynamic range of the image, and K1 and K2 are small constants; based on the SSIM value, the detail loss anomaly index DLI is calculated to measure the degree of detail loss due to image transmission delay and compression. The expression is: SSIM threshold is the preset SSIM threshold.

[0015] Preferably, in S2, after analyzing the fluctuation of the compression ratio of each frame of image, an image compression ratio fluctuation index is generated. The method for obtaining the image compression ratio fluctuation index is:

[0016] For each frame of image, calculate its compression ratio CR i , and record it. Assuming that the image sequence has N frames, the compression ratio data is {CR1, CR2, ..., CR N}; Calculate the frequency distribution of all compression ratios and obtain the CR of each compression ratio i The probability of occurrence in the entire data set p(CR i ); the calculation formula for frequency distribution is: Among them, the frequency (CR i ) is the compression ratio CR i The number of occurrences, N is the total number of frames of compression ratio data, and entropy is calculated through the probability distribution of compression ratio, expressed as: Among them, p(CR i ) is the compression ratio CR i The probability of n is , n is the number of different compression ratio values, H(CR) is the entropy of the compression ratio, and the image compression ratio fluctuation index is calculated. The expression is: CRFI = H(CR); where CRFI is the image compression ratio fluctuation index.

[0017] Preferably, in S4, the degree of image quality degradation is evaluated based on the impact of the delay on the real-time performance of the image and the image compression quality. According to the evaluation result, the degree of image quality degradation is divided into severe degradation and slight degradation, specifically:

[0018] The detail loss anomaly index and the image compression ratio fluctuation index are converted into comprehensive feature vectors, which are used as inputs of the machine learning model. The machine learning model uses the quality degradation value label predicted by each set of comprehensive feature vectors as the prediction target, and minimizes the sum of the prediction errors of the quality degradation value labels of all images as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the model training is stopped. The quality degradation value of the image is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0019] Preferably, the acquired image quality degradation value is compared with a reference threshold of the image quality degradation value set according to historical data. If the image quality degradation value is greater than or equal to the set reference threshold of the image quality degradation value, it indicates that the image quality degradation is serious. In this case, a serious image quality degradation signal is generated, and the image quality degradation is classified as serious degradation. If the image quality degradation value is less than the set reference threshold of the image quality degradation value, it indicates that the image quality degradation is slight. In this case, a slight image quality degradation signal is generated, and the image quality degradation is classified as slight degradation.

[0020] Preferably, in S5, for severe degradation, that is, the quality degradation value of the image generated within a fixed time period is greater than or equal to the set reference threshold of the image quality degradation value, the quality degradation values ​​generated in subsequent fixed time periods that are greater than or equal to the reference threshold of the image quality degradation value are collected, and a corresponding data set is established, the mean and standard deviation of the data set are calculated, and the accuracy of the image recognition algorithm within the fixed time period is predicted and warned.

[0021] Preferably, if the mean of the quality degradation values ​​in the data set is greater than or equal to the reference threshold of the mean of the quality degradation values, and the standard deviation of the quality degradation values ​​is less than the reference threshold of the standard deviation of the quality degradation values, a first-level warning signal is generated, and the image recognition algorithm needs to be adjusted immediately to increase the fault tolerance of image processing;

[0022] If the mean of the quality degradation values ​​is greater than or equal to the reference threshold of the mean of the quality degradation values, and the standard deviation of the quality degradation values ​​is greater than or equal to the reference threshold of the standard deviation of the quality degradation values, a level 2 warning signal is generated, and some algorithm optimization and parameter adjustment are required;

[0023] If the mean of the quality degradation values ​​is less than the reference threshold of the mean of the quality degradation values, and the standard deviation of the quality degradation values ​​is greater than or equal to the reference threshold of the standard deviation of the quality degradation values, a level 3 warning signal is generated, and the image transmission path needs to be optimized;

[0024] If the mean of the quality degradation values ​​is less than the reference threshold of the mean of the quality degradation values, and the standard deviation of the quality degradation values ​​is less than the reference threshold of the standard deviation of the quality degradation values, no warning signal is generated, and there is no need to adjust the image recognition algorithm, and the current operation continues.

[0025] The present invention also provides an airborne surface mine and unexploded ordnance detection system based on image recognition, comprising an image acquisition and signal monitoring module, a signal delay assessment module, an image compression quality analysis module, an image quality classification module, and an image enhancement and recognition adjustment module:

[0026] Image acquisition and signal monitoring module: The drone flies over the target area through a remote control command system. The high-resolution camera on board collects real-time image information of surface mines, unexploded ordnance, and their surroundings. The wireless communication signal strength between the drone and the ground receiver is measured and recorded in real time.

[0027] Signal Delay Assessment Module: Compares and analyzes signal strength fluctuations and image transmission delays to determine the extent to which delays interfere with image resolution and detail retention, and assesses the impact of delays on image real-time performance.

[0028] Image compression quality analysis module: If the delay is serious, the stored compression ratio information of each frame of the image during the image compression process is analyzed, and the difference between the compressed image and the original image is compared to evaluate the image compression quality;

[0029] Image quality classification module: This module evaluates the degree of image quality degradation based on the impact of latency on image real-time performance and image compression quality. Based on the evaluation results, the image quality degradation is classified as severe or slight.

[0030] Image enhancement and recognition adjustment module: For minor drops, the image quality is improved by locally enhancing the image, making the features of surface mines and unexploded ordnance clearer. For serious drops, the accuracy of the image recognition algorithm within a fixed time period is predicted and warned, and the image recognition algorithm is automatically adjusted based on the prediction results to improve the accuracy of surface mine and unexploded ordnance detection.

[0031] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0032] 1. This invention accurately assesses the impact of image quality by measuring the signal strength between the drone and the ground receiver in real time, combined with analysis of image transmission delay. It also quantifies the degree of image quality degradation using the detail loss anomaly index and the image compression ratio fluctuation index. Combined with predictions from machine learning models, it can dynamically determine whether image quality has degraded slightly or significantly, and implement local enhancement or adjust the image recognition algorithm based on the situation, thereby improving the accuracy and stability of surface mine and unexploded ordnance detection.

[0033] 2. This invention effectively addresses the uncertainties of wireless communication environments through precise control of image quality and algorithm adjustments, ensuring the provision of high-quality image information even in complex terrain and weather conditions. By comprehensively evaluating the impact of image compression quality and transmission delay on image real-time performance, and implementing multi-level adjustments through an early warning mechanism, this not only improves the reliability of the surface mine and unexploded ordnance detection system, but also effectively reduces image recognition errors, ensuring the efficient completion of detection missions. This innovative approach significantly enhances the adaptability and accuracy of drone-mounted image recognition systems in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0035] Figure 1 Flow chart of the method of the present invention.

[0036] Figure 2 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0038] Example 1, please refer to Figure 1 and Figure 2 As shown, the method for detecting surface mines and unexploded ordnance based on image recognition in this embodiment includes the following steps:

[0039] S1: The drone flies over the target area through a remote control command system. The drone uses its onboard high-resolution camera to collect real-time images of surface mines, unexploded ordnance, and their surroundings. It also measures and records the wireless communication signal strength between the drone and the ground receiver.

[0040] S2: Compare and analyze signal strength fluctuations and image transmission delays to determine the extent to which delays interfere with image resolution and detail preservation, and assess the impact of delays on image real-time performance.

[0041] S3: If the delay is severe, the stored compression ratio information of each frame of the image during the image compression process is analyzed, and the difference between the compressed image and the original image is compared to evaluate the image compression quality;

[0042] S4: Evaluate the degree of image quality degradation based on the impact of the delay on image real-time performance and image compression quality, and classify the image quality degradation into severe degradation and slight degradation based on the evaluation results;

[0043] S5: For minor drops, the image quality is improved by locally enhancing the image, making the features of surface mines and unexploded ordnance clearer. For severe drops, the accuracy of the image recognition algorithm within a fixed time period is predicted and warned, and the image recognition algorithm is automatically adjusted based on the prediction results to improve the accuracy of surface mine and unexploded ordnance detection.

[0044] In S1, the drone flies to the target area through a remote control command system. The drone uses its high-resolution camera to collect real-time image information of surface mines, unexploded ordnance, and their surroundings. The drone also measures and records the wireless communication signal strength between the drone and the ground receiver in real time. Specifically:

[0045] Drones are controlled via a remote control command system, which typically includes both automatic navigation and manual control capabilities for precise positioning and flight path planning. Over the target area, the drone will fly along a predetermined trajectory. Technologies such as GPS, an inertial navigation system (INS), and a visual navigation system (VIO) ensure precise positioning during flight. During flight, the drone's position, speed, and altitude are provided in real time, ensuring flight accuracy and coverage of the image acquisition area. High-resolution cameras capture real-time images of the target area, identifying potential areas of surface mines and unexploded ordnance. Positioning information and image data from these areas are recorded for subsequent processing.

[0046] The high-resolution camera onboard provides high-quality images, ensuring clear capture of details of surface mines and unexploded ordnance even at long distances. The camera automatically adjusts exposure, focus, and other imaging parameters based on flight altitude and speed to achieve optimal image quality. Captured images are typically stored in lossless formats such as RAW or TIFF to ensure image data integrity and provide high-quality raw data for subsequent image processing and analysis. The camera captures images in real time and transmits them to a ground receiver via wireless communication. This process is highly dependent on transmission bandwidth, latency, and image compression algorithms.

[0047] The wireless communication equipment (such as Wi-Fi, LTE, 5G, etc.) carried by the drone continuously monitors the signal strength between it and the ground receiver. The drone's built-in signal strength detection module collects and records signal strength data in real time. As the drone flies, the signal strength will be affected by terrain, weather, and obstacles (such as buildings, mountains, etc.), and may fluctuate. By measuring signal strength, the system can identify areas with weakened signals in real time and provide a basis for subsequent image transmission optimization. The signal strength data is associated with the collected image information, and the changes in signal strength at each moment are recorded. This provides data support for subsequent analysis of image transmission quality and its impact on signal strength fluctuations.

[0048] To accurately analyze the relationship between signal strength and image quality, the system must timestamp both image and signal strength data. This ensures that subsequent steps can accurately identify the correlation between signal strength fluctuations and image quality degradation. Signal strength data and image acquisition data must be synchronized in real time to facilitate detection of any interruptions or delays in image transmission. This real-time feedback allows operators to adjust the drone's flight path or communication strategy based on actual conditions.

[0049] S2: Compare and analyze signal strength fluctuations and image transmission delays to determine the extent to which delays interfere with image resolution and detail preservation, and assess the impact of delays on image real-time performance.

[0050] Drones continuously monitor signal strength during flight using onboard wireless communication equipment (such as Wi-Fi, 4G, 5G, or dedicated drone communication protocols). This signal strength data is recorded in real time and stored synchronously with the image data. By analyzing changes in wireless signal strength, it is possible to determine whether the signal strength fluctuates dramatically or remains within a relatively stable range. Excessive fluctuations can lead to decreased image transmission quality or even packet loss.

[0051] Image transmission latency refers to the time delay between the drone's camera capturing an image and its arrival at the ground receiver. This latency is dependent on various factors, including signal strength, flight distance, wireless bandwidth, and image compression method. The system records the transmission time of each image frame and calculates average and maximum latency. High-resolution images require a large bandwidth, especially during long-distance flights and when transmitting high data volumes. Insufficient bandwidth can increase image transmission latency. By comparing the relationship between bandwidth and latency, we can identify the key factors influencing latency.

[0052] When there is a delay in image transmission, especially when the signal is unstable or the bandwidth is insufficient, the image may need to be compressed or the resolution may need to be reduced to accommodate the bandwidth requirements of the transmission. The reduction in resolution directly affects the detail of the image and may result in the loss of details of surface mines and unexploded ordnance, thus affecting subsequent identification and classification. Analysis method: The image compression ratio and resolution changes under different delay conditions can be calculated to compare the differences in details between the original image and the compressed image. For example, edge detection, contrast analysis and other techniques can be used to quantify the degree of resolution loss. If the delay is too long, the image may be distorted or blurred due to network congestion or signal attenuation, especially the outlines, details and color changes of surface mines and unexploded ordnance, which may lead to misjudgment or ineffective identification.

[0053] Image transmission delay not only affects resolution but can also lead to loss of image detail during the image compression process. This is especially true for targets requiring precise identification, such as surface mines and unexploded ordnance. Any loss of detail can affect the algorithm's recognition accuracy. The effect of image transmission delay on detail loss in the compressed image and the original image can be compared. For example, image similarity metrics (such as the structural similarity index (SSIM)) can be used to measure the degree of detail preservation, particularly for details such as the target's outline, color, and texture. Excessive detail loss due to delay can affect the accurate identification of surface mines and unexploded ordnance, increasing the risk of misidentification and missed detections.

[0054] Real-time image transmission is crucial when drones are performing missions, especially in dynamically changing environments, where the timeliness of images is directly related to subsequent decision-making and disposal. For example, real-time image recognition can quickly determine the location, type, and hazard of surface mines and unexploded ordnance, helping to determine whether further action is needed. Real-time impact analysis: By comparing the update frequency and timeliness of images under different delay conditions, it is possible to assess whether the delay exceeds the system's tolerance. If image updates lag too far, the information may become outdated and fail to reflect changes in the current environment, thus affecting mission execution.

[0055] By comparing the effect of image transmission delay on the detail loss of the compressed image and the original image, a detail loss anomaly index is generated to evaluate the impact of delay on the real-time performance of the image. The method for obtaining the detail loss anomaly index is as follows:

[0056] Obtain the original image and the compressed image: The original image is an image captured in real time from the drone, untransmitted or uncompressed. The compressed image is an image that has been transmitted or compressed (which may be affected by network latency and the compression algorithm). Match the sizes of the two images to ensure they have the same resolution (if the images are of different sizes, interpolation can be performed). Grayscale both images (if they are color images, convert them to grayscale for calculation).

[0057] The structural similarity index SSIM is used to measure the similarity between two images, taking into account three aspects: brightness, contrast, and structure. The formula is: Where x and y are the corresponding pixel values ​​of the original image and the compressed image, μ x and μ y is the local mean (brightness information) of the original image and the compressed image, that is, the average value of the image block, and is the local variance (contrast information) between the original image and the compressed image, indicating the degree of brightness change in the local area. xyis the local covariance (structural information) of the original image and the compressed image, indicating the degree of similarity between the two images. C1 and C2 are constants used to avoid the denominator being zero. Usually, C1 = K1L) 2 and C2=(K2L) 2 , where L is the dynamic range of the image (e.g. 255 for an 8-bit image), K1 and K2 are small constants, usually K1 = 0.01 and K2 = 0.03.

[0058] According to the SSIM value, the detail loss anomaly index DLI is calculated to measure the degree of detail loss caused by image transmission delay and compression. The expression is: SSIM threshold This is the preset SSIM threshold used to judge image quality. It can usually be set to 0.7. If the SSIM is lower than this value, it means that the image details are severely lost.

[0059] The larger the Detail Loss Anomaly Index (DLI), the more severe the impact of latency on image real-time performance. This is because the DLI calculation method indicates that when significant detail is lost during image compression or transmission, the SSIM value will be lower. The larger the DLI value, the more significant the information loss during the compression process, the poorer the detail retention, and the significant degradation of image quality. This is usually related to factors such as long transmission delays, signal packet loss, or excessive compression, resulting in unclear images and blurred details, thereby affecting the accuracy of image recognition algorithms. Therefore, a large DLI value indicates that image quality has been significantly affected, and the real-time performance and accuracy of the system may be seriously disturbed.

[0060] The smaller the Detail Loss Anomaly Index (DLI), the less impact the delay has on the real-time performance of the image. A smaller DLI value indicates that the image retains more details during transmission, with less quality loss due to compression or delay. Typically, an SSIM value close to 1 indicates that there is little loss of image detail and structural information. At this point, even with a certain amount of signal delay or compression, the image still maintains a high quality, enabling relatively accurate target detection and recognition. Therefore, a smaller DLI value generally means that the real-time performance and accuracy of the image are less affected, and the system can perform image recognition tasks more reliably.

[0061] S3: If the delay is severe, the stored compression ratio information of each frame of the image during the image compression process is analyzed, and the difference between the compressed image and the original image is compared to evaluate the compression quality of the image.

[0062] The compression ratio (CR) is the ratio of the size of the image before compression to the size of the image after compression, reflecting the degree of data reduction during the compression process. Its formula is: Original Image Size is the size of the original image, typically measured in bytes. Compressed Image Size is the size of the compressed image, also measured in bytes. A higher compression ratio means a smaller compressed image file and better compression efficiency, but this may result in a greater loss of detail. Conversely, a lower compression ratio means less compression, better detail preservation, but the image file size remains larger.

[0063] For each frame of image captured by the drone, the image size before and after compression must be measured, and the compression ratio must be calculated. The specific steps are as follows:

[0064] Get the original size of each frame of image: measure the file size of each frame of original image and record the number of bytes of the original image.

[0065] Compressed image: Each frame of the image is compressed, usually using some image compression algorithm (such as JPEG, JPEG2000, HEVC, etc.).

[0066] Get the size of the compressed image: After compression, record the size of the image file (Bytes).

[0067] Calculate the compression ratio: Use the above formula to calculate the compression ratio of each frame of image.

[0068] The image compression ratio directly affects the quality of the compressed image. The higher the compression ratio, the more the image is compressed, and the more image details may be lost. Generally speaking:

[0069] High compression ratios (e.g., compression ratios greater than 10:1) typically result in significant quality loss, especially in details and textures; the image may appear blurry or distorted.

[0070] A low compression ratio (for example, a compression ratio close to 1:1) retains more details, but the compression efficiency is poor and the image file size is larger.

[0071] By analyzing the compression ratio of each frame, we can infer the image quality. For example, if the compression ratio is too high, it may cause loss of image details, blurred textures, color distortion and other problems, affecting the subsequent image recognition process.

[0072] After analyzing the compression ratio fluctuation of each frame of image, the image compression ratio fluctuation index is generated to evaluate the image compression quality. The method for obtaining the image compression ratio fluctuation index is as follows:

[0073] For each frame of image, calculate its compression ratio CR i, and record it. Assuming that the image sequence has N frames, the compression ratio data is {CR1, CR2, ..., CR N}; Calculate the frequency distribution of all compression ratios and obtain the CR of each compression ratio i The probability of occurrence in the entire data set p(CR i ); the calculation formula for frequency distribution is: Among them, the frequency (CR i ) is the compression ratio CR i The number of occurrences, N is the total number of frames of compression ratio data, and entropy is calculated through the probability distribution of compression ratio, expressed as: Among them, p(CR i ) is the compression ratio CR i The probability of compression ratio is , n is the number of different compression ratio values ​​(i.e., the number of discrete values ​​of compression ratio), H(CR) is the entropy of compression ratio, which indicates the uncertainty or volatility of compression ratio. The image compression ratio fluctuation index is calculated as follows: CRFI = H9CR); where CRFI is the image compression ratio fluctuation index.

[0074] A larger compression ratio fluctuation index indicates a more unstable image compression quality, potentially leading to significant image quality variations. A high compression ratio fluctuation index indicates significant variations in the compression ratio between image frames, typically due to poor adaptation to image content during the compression process or an unbalanced dynamic adjustment strategy. A high fluctuation index can result in some frames experiencing excessively low compression quality (severe loss of detail) while others may be over-retained (occupying more storage space). This instability can interfere with subsequent image recognition or analysis tasks, particularly in scenarios requiring consistent image quality.

[0075] A smaller image compression ratio fluctuation index indicates more stable image compression quality and more consistent quality across frames. A smaller fluctuation index means smaller variations in the compression ratio, indicating greater adaptability of the compression algorithm across frames, balancing image quality and storage requirements. This stability helps maintain consistency across image sequences, reducing recognition errors or visual discomfort caused by inter-frame quality variations, while also improving system compression efficiency and reliability.

[0076] S4: Evaluate the degree of image quality degradation based on the impact of the delay on the real-time performance of the image and the image compression quality. Based on the evaluation result, classify the degree of image quality degradation into severe degradation and slight degradation.

[0077] The detail loss anomaly index and the image compression ratio fluctuation index are converted into comprehensive feature vectors, which are used as inputs of the machine learning model. The machine learning model uses the quality degradation value label predicted by each set of comprehensive feature vectors as the prediction target, and minimizes the sum of the prediction errors of the quality degradation value labels of all images as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the model training is stopped. The quality degradation value of the image is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0078] The method for obtaining the image quality degradation value is as follows: from the comprehensive feature vector training data of the trained machine learning model, the corresponding function expression is obtained: R = F9DLI, CRFI); where F is the output function of the model, DLI is the detail loss anomaly index, CRFI is the image compression ratio fluctuation index, and R is the image quality degradation value.

[0079] The acquired image quality degradation value is compared with a reference threshold of the image quality degradation value set according to historical data. If the image quality degradation value is greater than or equal to the set reference threshold of the image quality degradation value, it indicates that the image quality degradation is serious. In this case, a serious image quality degradation signal is generated, and the image quality degradation is classified as serious degradation. If the image quality degradation value is less than the set reference threshold of the image quality degradation value, it indicates that the image quality degradation is slight. In this case, a slight image quality degradation signal is generated, and the image quality degradation is classified as slight degradation.

[0080] S5: For minor drops, the image quality is improved by locally enhancing the image, making the features of surface mines and unexploded ordnance clearer. For severe drops, the accuracy of the image recognition algorithm within a fixed time period is predicted and warned, and the image recognition algorithm is automatically adjusted based on the prediction results to improve the accuracy of surface mine and unexploded ordnance detection.

[0081] The goal of local enhancement is to make the details of surface mines and unexploded ordnance clearer by optimizing key areas in the image, thereby improving their detectability and accuracy in subsequent image recognition. For slightly degraded images, details can be restored by enhancing contrast, brightness, and sharpening, particularly in areas lost during compression or transmission.

[0082] The local enhancement method first needs to identify the important areas in the image, that is, the areas where surface mines and unexploded ordnance may be present. This can be achieved by:

[0083] Selection based on region of interest (ROI): Usually, target detection or image segmentation technology is used to roughly locate areas that may contain surface mines and unexploded ordnance (such as circular or rectangular candidate boxes).

[0084] Edge detection: Edge detection algorithms (such as Canny edge detection and Sobel operator) can be used to determine the edges of objects in the image, thereby selecting areas that may contain surface mines and unexploded ordnance.

[0085] Deep learning prediction: Use convolutional neural networks (CNN) or other object detection algorithms (such as YOLO or Faster R-CNN) to automatically identify target objects in images and determine key areas.

[0086] According to the type of image quality degradation, select the appropriate local enhancement algorithm to improve image clarity. Common local enhancement algorithms include:

[0087] Histogram equalization can enhance the contrast of an image, especially in low-contrast areas. By enhancing the contrast of local areas, the details of surface mines and unexploded ordnance can be made more prominent.

[0088] Adaptive Histogram Equalization: This method can perform histogram equalization based on the image features of the local area to avoid excessive contrast enhancement of the overall image. It is suitable for local enhancement.

[0089] Local sharpening mainly helps restore details lost due to compression and transmission by enhancing edge details of the image.

[0090] Edge detection algorithms such as the Laplacian operator or the Sobel operator are used, combined with weighted enhancement of local areas, to enhance the outlines and details of surface mines and unexploded ordnance.

[0091] CLAHE (Contrast Limited Adaptive Histogram Equalization) is a local enhancement algorithm that effectively improves contrast in local areas while avoiding over-enhancement. It is suitable for restoring image features even when detail is lost. CLAHE divides the image into blocks and performs histogram equalization on each block, while limiting the contrast improvement within each block to avoid noise enhancement.

[0092] Local brightness enhancement compensates for dimly lit areas, making image details more visible. Adaptive brightness adjustment can better restore the visible features of surface mines and unexploded ordnance. Using a local filter to adjust brightness makes areas of surface mines and unexploded ordnance more recognizable. After completing local enhancement, the results need to be evaluated. You can verify the image enhancement results using the following methods:

[0093] By comparing the contrast and clarity of the images before and after enhancement, we can verify whether the image details have been enhanced. We then use an image recognition model to detect the enhanced images and evaluate the recognition rate of surface mines and unexploded ordnance.

[0094] In the event of a severe decline, the accuracy of the image recognition algorithm within a fixed time period is predicted and warned, and the image recognition algorithm is automatically adjusted based on the prediction results to improve the accuracy of surface mine and unexploded ordnance detection. Specifically:

[0095] For severe degradation, that is, the quality degradation value of the image generated within a fixed time period is greater than or equal to the set reference threshold of the image quality degradation value, the quality degradation values ​​generated in the subsequent fixed time period that are greater than or equal to the reference threshold of the image quality degradation value are collected, and a corresponding data set is established, and the mean and standard deviation of the data set are calculated to predict and warn the accuracy of the image recognition algorithm within the fixed time period.

[0096] If the mean of the quality degradation values ​​in the data set is greater than or equal to the reference threshold of the mean of the quality degradation values, and the standard deviation of the quality degradation values ​​is less than the reference threshold of the standard deviation of the quality degradation values, a level one warning signal is generated. The level one warning signal is the most serious, meaning that the change in the quality degradation is small but the persistence is strong. The image recognition algorithm needs to be adjusted immediately, and it may be necessary to enable redundant detection or increase the fault tolerance of image processing.

[0097] If the mean of the quality degradation values ​​is greater than or equal to the reference threshold of the mean of the quality degradation values, and the standard deviation of the quality degradation values ​​is greater than or equal to the reference threshold of the standard deviation of the quality degradation values, a secondary warning signal is generated. The secondary warning signal indicates that the quality degradation fluctuates within a certain range, and some algorithm optimization is required, such as adjusting parameters or enhancing the diversity of the data set.

[0098] If the mean of the quality degradation values ​​is less than the reference threshold of the mean of the quality degradation values, and the standard deviation of the quality degradation values ​​is greater than or equal to the reference threshold of the standard deviation of the quality degradation values, a level 3 warning signal is generated. The level 3 warning signal indicates that the image quality degradation is highly volatile, but the overall degradation is not serious. This may indicate network interference or short-term performance fluctuations, and it is necessary to optimize the image transmission path or perform image enhancement.

[0099] If the mean of the quality degradation values ​​is less than the reference threshold of the mean of the quality degradation values, and the standard deviation of the quality degradation values ​​is less than the reference threshold of the standard deviation of the quality degradation values, no warning signal is generated. No warning signal indicates that the image quality is good, and there is no need to adjust the image recognition algorithm. The current operation can continue.

[0100] It should be noted here that the importance of the first-level warning signal is greater than that of the second-level warning signal, and the importance of the second-level warning signal is greater than that of the third-level warning signal. Relevant personnel can take corresponding treatment measures according to different warning signal levels.

[0101] Based on different warning signal levels, different measures are taken to adjust the image recognition algorithm to improve the accuracy of surface mine and unexploded ordnance detection:

[0102] Level 1 warning: Global adjustments to the image recognition algorithm may be made, such as adding redundant detection, increasing image preprocessing steps, and improving image resolution. More complex deep learning models may be enabled to address poor image quality.

[0103] Level 2 warning: Perform local optimization adjustments, such as adjusting model parameters or enhancing the diversity of the dataset, to adapt to fluctuations in image quality.

[0104] Level 3 warning: Improve recognition accuracy by improving the transmission path, enhancing the signal, or performing local image enhancement, while keeping the image quality within an acceptable range.

[0105] In this embodiment, a drone-mounted high-resolution camera captures real-time images of surface mines, unexploded ordnance, and the surrounding environment. Combined with real-time measurement of wireless communication signal strength, the system analyzes image transmission delay and compression quality. When signal delay significantly impacts image real-time performance and detail preservation, the system assesses the image compression ratio and quality to determine the degree of image quality degradation, categorizing it as severe or mild. In the case of mild degradation, image quality is enhanced through local enhancement, clarifying the features of surface mines and unexploded ordnance. In the case of severe degradation, the system predicts the accuracy of the image recognition algorithm, providing timely warnings and automatically adjusting the recognition algorithm to improve the accuracy of surface mine and unexploded ordnance detection.

[0106] Example 2: The airborne surface mine and unexploded ordnance detection system based on image recognition described in this example includes an image acquisition and signal monitoring module, a signal delay assessment module, an image compression quality analysis module, an image quality classification module, and an image enhancement and recognition adjustment module:

[0107] Image acquisition and signal monitoring module: The drone flies over the target area through a remote control command system. The high-resolution camera on board collects real-time image information of surface mines, unexploded ordnance, and their surroundings. The wireless communication signal strength between the drone and the ground receiver is measured and recorded in real time.

[0108] Signal Delay Assessment Module: Compares and analyzes signal strength fluctuations and image transmission delays to determine the extent to which delays interfere with image resolution and detail retention, and assesses the impact of delays on image real-time performance.

[0109] Image compression quality analysis module: If the delay is serious, the stored compression ratio information of each frame of the image during the image compression process is analyzed, and the difference between the compressed image and the original image is compared to evaluate the image compression quality;

[0110] Image quality classification module: This module evaluates the degree of image quality degradation based on the impact of latency on image real-time performance and image compression quality. Based on the evaluation results, the image quality degradation is classified as severe or slight.

[0111] Image enhancement and recognition adjustment module: For minor drops, the image quality is improved by locally enhancing the image, making the features of surface mines and unexploded ordnance clearer. For serious drops, the accuracy of the image recognition algorithm within a fixed time period is predicted and warned, and the image recognition algorithm is automatically adjusted based on the prediction results to improve the accuracy of surface mine and unexploded ordnance detection.

[0112] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0113] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0114] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0115] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. An airborne surface mine and unexploded ordnance detection method based on image recognition, characterized in that: The following steps are involved: S1: The UAV flies to the target area through the remote control command system, collects image information of surface mines, unexploded ordnance and their surroundings in real time through the high-resolution camera on board, and measures and records the wireless communication signal strength between the UAV and the ground receiver in real time; S2: Compare and analyze the signal strength fluctuation and the delay of image transmission to determine the interference degree of delay on image resolution and detail retention, and evaluate the impact of delay on image real-time performance; S3: If the delay has a serious impact, the stored compression ratio information of each frame of the image during the image compression process is analyzed, the difference between the compressed image and the original image is compared, and the compression quality of the image is evaluated; S4: evaluating the degree of image quality degradation according to the impact of the delay on the real-time performance of the image and the image compression quality, and classifying the image quality degradation into severe degradation and slight degradation according to the evaluation result; S5: For slight descent, the image quality is improved by locally enhancing the image, making the features of surface mines and unexploded ordnance clearer; For severe declines, the accuracy of the image recognition algorithm within a fixed time period is predicted and warned, and the image recognition algorithm is automatically adjusted based on the prediction results to improve the accuracy of surface mines and unexploded ordnance detection.

2. The method for detecting surface mines and unexploded ordnance based on image recognition according to claim 1, characterized in that: In S2, the detail loss anomaly index is generated by comparing the detail loss degree of the compressed image and the original image due to the image transmission delay. The method for obtaining the detail loss anomaly index is as follows: Get the original image and the compressed image, match the size of the two images, and perform grayscale processing on the two images. The structural similarity index SSIM is used to measure the similarity between the two images, taking into account three aspects: brightness, contrast, and structure. The formula is: Where x and y are the corresponding pixel values ​​of the original image and the compressed image, μx and μy are the local means of the original image and the compressed image, that is, the average values ​​of the image blocks. and is the local variance of the original image and the compressed image, indicating the degree of brightness change in the local area, σxy is the local covariance of the original image and the compressed image, indicating the degree of similarity between the two images; C1 and C2 are constants used to avoid the denominator being zero, where L is the dynamic range of the image, K1 and K2 are small constants; according to the SSIM value, the detail loss anomaly index DLI is calculated to measure the degree of detail loss caused by image transmission delay and compression, and the expression is: SSIM threshold is the preset SSIM threshold.

3. The method for detecting surface mines and unexploded ordnance based on image recognition according to claim 2, characterized in that: In S2, the fluctuation of the compression ratio of each frame of the image is analyzed to generate an image compression ratio fluctuation index. The method for obtaining the image compression ratio fluctuation index is as follows: For each frame of image, calculate its compression ratio CR i , and record it. Assuming that the image sequence has N frames, the compression ratio data is {CR1, CR2, …, CR N }; Calculate the frequency distribution of all compression ratios and obtain the CR of each compression ratio i The probability of occurrence in the entire data set p(CR i ); the frequency distribution is calculated as: Among them, the frequency (CR i ) is the compression ratio CR i The number of occurrences, N is the total number of frames of compression ratio data, and entropy is calculated through the probability distribution of compression ratio, expressed as: Among them, p9CR i ) is the compression ratio CR i The probability of , n is the number of different compression ratio values, H(CR) is the entropy of the compression ratio, and the image compression ratio fluctuation index is calculated. The expression is: CRFI=H(CR); where CRFI is the image compression ratio fluctuation index.

4. The method for detecting surface mines and unexploded ordnance based on image recognition according to claim 3 is characterized in that: In S4, the degree of image quality degradation is evaluated according to the impact of the delay on the real-time performance of the image and the image compression quality. According to the evaluation result, the degree of image quality degradation is divided into severe degradation and slight degradation, specifically: The detail loss anomaly index and the image compression ratio fluctuation index are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as the input of the machine learning model. The machine learning model takes the quality degradation value label of the image predicted by each group of comprehensive feature vectors as the prediction target, and takes minimizing the sum of prediction errors of the quality degradation value labels of all images as the training target. The machine learning model is trained until the sum of prediction errors converges, and the model training is stopped. The quality degradation value of the image is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

5. The method for detecting surface mines and unexploded ordnance based on image recognition according to claim 4, characterized in that: The acquired image quality degradation value is compared with the reference threshold of the image quality degradation value set according to historical data. If the image quality degradation value is greater than or equal to the set reference threshold of the image quality degradation value, it means that the image quality degradation is serious. In this case, a serious image quality degradation signal is generated, and the image quality degradation is classified as serious degradation. If the image quality degradation value is less than the set reference threshold of the image quality degradation value, it means that the image quality degradation is slight. In this case, a slight image quality degradation signal is generated, and the image quality degradation is classified as slight degradation.

6. The method for detecting surface mines and unexploded ordnance based on image recognition according to claim 1, characterized in that: In S5, for severe degradation, that is, the quality degradation value of the image generated within a fixed time period is greater than or equal to the set reference threshold of the image quality degradation value, the quality degradation values ​​generated in the subsequent fixed time period that are greater than or equal to the reference threshold of the image quality degradation value are collected, and a corresponding data set is established, and the mean and standard deviation of the data set are calculated to predict and warn the accuracy of the image recognition algorithm within the fixed time period.

7. The method for detecting surface mines and unexploded ordnance based on image recognition according to claim 6, characterized in that: If the mean value of the quality degradation degree in the data set is greater than or equal to the reference threshold value of the mean value of the quality degradation degree, and the standard deviation of the quality degradation degree is less than the reference threshold value of the standard deviation of the quality degradation degree, a first-level warning signal is generated at this time, and the image recognition algorithm needs to be adjusted immediately to increase the fault tolerance of image processing; If the mean value of the quality degradation degree is greater than or equal to the reference threshold value of the mean value of the quality degradation degree, and the standard deviation of the quality degradation degree is greater than or equal to the reference threshold value of the standard deviation of the quality degradation degree, a secondary warning signal is generated at this time, and some algorithm optimization and parameter adjustment are required; If the mean value of the quality degradation degree is less than the reference threshold value of the mean value of the quality degradation degree, and the standard deviation of the quality degradation degree is greater than or equal to the reference threshold value of the standard deviation of the quality degradation degree, a third-level warning signal is generated at this time, and the image transmission path needs to be optimized; If the mean of the quality degradation values ​​is less than the reference threshold of the mean of the quality degradation values, and the standard deviation of the quality degradation values ​​is less than the reference threshold of the standard deviation of the quality degradation values, no warning signal is generated, and there is no need to adjust the image recognition algorithm, and the current operation continues.

8. An airborne surface mine and unexploded ordnance detection system based on image recognition, used to implement the airborne surface mine and unexploded ordnance detection method based on image recognition as claimed in any one of claims 1 to 7, characterized in that: It includes image acquisition and signal monitoring module, signal delay evaluation module, image compression quality analysis module, image quality classification module and image enhancement and recognition adjustment module: Image acquisition and signal monitoring module: The UAV flies to the target area through the remote control command system, collects image information of surface mines, unexploded ordnance and their surroundings in real time through the high-resolution camera on board, and measures and records the wireless communication signal strength between the UAV and the ground receiver in real time; Signal delay evaluation module: compares and analyzes the signal strength fluctuation and the delay of image transmission, determines the interference degree of delay on image resolution and detail retention, and evaluates the impact of delay on image real-time performance; Image compression quality analysis module: If the delay is serious, the storage compression ratio information of each frame of the image during the image compression process is analyzed, the difference between the compressed image and the original image is compared, and the compression quality of the image is evaluated; Image quality classification module: evaluates the degree of image quality degradation according to the impact of delay on image real-time performance and image compression quality, and classifies the image quality degradation into severe degradation and slight degradation according to the evaluation results; Image enhancement and recognition adjustment module: For slight drops, the image quality is improved by locally enhancing the image, making the features of surface mines and unexploded ordnance clearer; For severe declines, the accuracy of the image recognition algorithm within a fixed time period is predicted and warned, and the image recognition algorithm is automatically adjusted based on the prediction results to improve the accuracy of surface mines and unexploded ordnance detection.

Images