A computer vision-based garbage detection system and method

Abstract

The present application relates to the field of visual processing, and more particularly to a garbage detection system and method based on computer vision, the garbage detection system comprising, a work detection module, which is used to determine whether to perform risk analysis on the garbage detection work process according to the work interference trend index periodically; an identification analysis module, which is used to determine whether to replace dynamic disturbance analysis when performing environmental interference analysis on the detection execution target based on the risk disturbance state in which the detection execution target is located; a dynamic disturbance evaluation module, which is used to determine the interference response evaluation parameters of the point disturbance space corresponding to the relevant interference point position of the detection execution target to perform review work detection; and an interference disturbance evaluation module, which is used to determine whether to perform detection on the interference floating index of the distribution analysis space according to the interference floating correlation parameters to perform review work detection on the distribution analysis space. The present application improves the overall visual analysis efficiency and quality of underwater garbage detection.

Description

Technical Field

[0001] This invention relates to the field of visual processing, and more particularly to a waste detection system and method based on computer vision. Background Technology

[0002] With the increasing urgency of marine environmental protection and the continuous development of underwater robotics technology, autonomous underwater vehicles (AUVs) and remotely operated underwater vehicles (ROVs) are gradually replacing manual diving operations. However, in actual underwater operations, there are numerous interfering factors that cause fluctuations in visual recognition quality. These include dynamic changes in water turbidity and sunlight, as well as self-induced turbidity caused by disturbances from the cooperating thrusters. Therefore, fixed detection schemes are difficult to effectively adapt to the real-time water conditions during actual operations, limiting the efficiency and long-term operational stability of underwater debris detection tasks. Thus, how to optimize and verify the underwater debris detection process in a timely manner, taking into account the dynamically changing environmental conditions during underwater operations, to ensure the reliability of underwater debris detection and the continuity of the task, is a problem that urgently needs to be solved by those skilled in the art.

[0003] Chinese patent application publication number CN121052480A discloses an intelligent method for detecting and collecting garbage in complex waters based on multimodal perception. This method acquires image perception data and semantic environment data from complex waters, preprocesses them, and performs feature extraction and semantic encoding. Then, it uses multimodal feature alignment and fusion, feature extraction, and multi-task detection to obtain garbage detection results. This effectively achieves deep fusion of semantic priors and visual features, improving the accuracy and robustness of garbage identification in complex water environments. Next, it assesses task importance to construct a garbage processing priority queue and builds a path dynamic planning strategy that coordinates energy consumption constraints and task priorities for path generation. This reduces navigation energy consumption while considering task benefits, thus achieving garbage detection and collection in complex waters. However, the above scheme has the following drawbacks: it fails to promptly optimize the underwater garbage detection process based on the actual impact of water state changes on the visual detection process, resulting in poor overall visual analysis efficiency and quality for underwater garbage detection. Summary of the Invention

[0004] To address this issue, the present invention provides a computer vision-based garbage detection system and method to overcome the problem in the prior art that it fails to promptly optimize the underwater garbage detection process based on the actual visual detection process being affected by changes in water conditions, resulting in poor overall visual analysis efficiency and quality for underwater garbage detection.

[0005] To achieve the above objectives, the present invention provides a computer vision-based waste detection system, comprising: The operation detection module is used to periodically determine whether to conduct risk analysis for the waste detection operation process based on the operation interference trend index of the detection execution target; The identification and analysis module, which is connected to the operation detection module, is used to determine whether to switch to dynamic disturbance analysis when performing environmental interference analysis on the target based on the risk disturbance state of the target. The risk disturbance state is determined based on the interference distribution correlation index and the correlation interference difference index of the target. A dynamic disturbance assessment module, which is connected to the identification and analysis module, is used to perform a verification operation based on the interference response assessment parameters of the disturbance space of the relevant interference points corresponding to the detection target. The interference response assessment parameters are determined based on the relevant assessment difference index and the time-series correlation assessment index, or by referring to the relevant difference index and the time-series correlation difference index. An interference disturbance assessment module, which is connected to the identification and analysis module, is used to determine whether to detect the interference fluctuation index of the target for the evaluation execution cycle based on the interference fluctuation correlation parameters, so as to perform a verification operation detection on the distribution analysis space of the target for the evaluation execution cycle.

[0006] Furthermore, the identification and analysis module responds to the analysis execution conditions and performs risk analysis on the waste detection operation process targeting the detection execution objective; The analysis execution condition is that the operation interference trend index determined by the evaluation execution cycle for the detection execution target is greater than the preset operation interference trend index. The operation interference trend index is determined based on the execution interference evaluation index determined by each interference evaluation cycle for the detection execution target within the interference evaluation stage.

[0007] Furthermore, the risk disturbance state includes a static risk disturbance state and a dynamic risk disturbance state; The detection execution target under the static risk disturbance state is the detection execution target where the interference distribution correlation index is less than or equal to the preset interference distribution correlation index and the correlation interference difference index is less than or equal to the preset correlation interference difference index; The detection execution target under the dynamic risk disturbance state is the detection execution target where the interference distribution correlation index is greater than the preset interference distribution correlation index or the correlation interference difference index is greater than the preset correlation interference difference index.

[0008] Furthermore, the dynamic disturbance assessment module performs dynamic disturbance analysis on the detection execution target that is in the dynamic risk disturbance state; The setting method for the interference response evaluation parameters of each point disturbance space of the detection execution target is determined based on the interference-related dispersion parameters and interference-related overlap parameters.

[0009] Furthermore, the dynamic disturbance assessment module determines the interference response assessment parameters of the disturbance space of the relevant interference points where the dispersed radiation target exists, based on the relevant assessment difference index and the time-series correlation assessment index. The dynamic disturbance assessment module determines the interference response assessment parameters of the disturbance space corresponding to the relevant interference points where the overlapping radiation target exists, based on the reference correlation difference index of each relevant interference point. Based on the temporal correlation difference index, the interference response evaluation parameters of the point disturbance space corresponding to the relevant interference points with temporal correlation difference indices that are less than or equal to the preset temporal correlation difference index for overlapping radiation targets are reduced and adjusted.

[0010] Furthermore, the dynamic disturbance assessment module performs a verification operation based on the verification interval parameter to detect the disturbance space of the relevant interference points corresponding to the detection target. The verification interval parameter of any point disturbance space is positively correlated with the corresponding interference response evaluation parameter of the point disturbance space.

[0011] Furthermore, the interference disturbance assessment module performs environmental interference analysis on the detection execution target in the static risk disturbance state; The interference floating correlation parameters are determined based on the interference floating parameters of each image acquisition point.

[0012] Furthermore, the interference disturbance assessment module detects the interference float index in the distribution analysis space where the interference float correlation parameter is greater than the preset interference float correlation parameter; The interference floating index is determined based on the illumination index floating parameters of each image acquisition point.

[0013] Furthermore, the interference disturbance assessment module performs a verification operation based on a stereo vision verification method to detect the distribution analysis space of the detection execution target where the interference fluctuation index is greater than the preset interference fluctuation index; The verification perspective parameters are determined based on the interference floating correlation parameters of the distribution analysis space corresponding to the detection execution target.

[0014] This invention also provides a computer vision-based garbage detection method, comprising: Periodically determine whether to conduct risk analysis for the waste detection process based on the operational interference trend index of the detection target; When conducting risk identification analysis, the risk disturbance state is determined based on the interference distribution correlation index and the relevant interference difference index of the detection target, and the environmental interference analysis for the detection target is changed to dynamic disturbance analysis based on the risk disturbance state of the detection target. When performing dynamic disturbance analysis, the interference response evaluation parameters of the disturbance space of the relevant interference points corresponding to the detection target are determined based on the relevant evaluation difference index and the time-series correlation evaluation index, or by referring to the relevant difference index and the time-series correlation difference index, so as to perform the verification operation detection. When performing environmental interference analysis, the interference float correlation parameters are used to determine whether to detect the interference float index of the target for evaluation of the distribution analysis space of the execution cycle, so as to perform a verification operation for the target for evaluation of the distribution analysis space of the execution cycle.

[0015] Compared with existing technologies, the advantages of this invention lie in its ability to quantify the risk of image quality deterioration over time by periodically monitoring and detecting the operational interference trend index of the target. This triggers subsequent disturbance analysis and review decisions, achieving a fundamental shift from passively responding to sudden drops in image quality to proactive early warning and strategy adaptation. This invention effectively avoids the prolonged operational interruptions and confidence collapse caused by passively remediating only after detection failure, significantly improving the reliability and analysis efficiency of underwater debris detection in complex and dynamic environments.

[0016] Furthermore, this invention accurately determines the risk disturbance state of the target being detected based on the interference distribution correlation index and the correlation interference difference index. The interference distribution correlation index quantifies the potential risk of spatial interference from surrounding collaborative equipment, while the correlation interference difference index quantifies the immediate deviation of the target's image quality from external influences. This categorizes risk states into two types, ensuring that the analysis of interference encountered during image acquisition aligns with reality. This invention avoids indiscriminate, computationally intensive full-scale analysis of all targets being detected, ensuring timely response in high-risk scenarios while minimizing unnecessary computational resource consumption.

[0017] Furthermore, this invention, for detection targets under dynamic risk disturbance, characterizes the spatial distribution discreteness of each relevant interference point and the temporal coupling degree of the interference source through interferometric correlation dispersion parameters and interferometric correlation overlap parameters. For dispersed radiation targets, interferometric response evaluation parameters of the point disturbance space are constructed based on the correlation evaluation difference index and the temporal correlation evaluation index, accurately quantifying the independent impact of isolated interference sources on local image quality. For overlapping radiation targets, the reference correlation difference index of each interferometric target is fused, and the temporal correlation difference index is introduced to correct the highly superimposed interference effect. This invention realizes the effectiveness of the analysis process for detection targets in underwater multi-machine cooperative operations, enhancing the reliability of the determined subsequent verification strategy.

[0018] Furthermore, this invention addresses occasional interference caused by changes in sunlight under static risk conditions by quantifying the correlation between image quality and illumination changes through interferometric floating correlation parameters. Based on this, the number of viewing angles for stereo vision verification is dynamically adjusted; the stronger the correlation, the richer the verification viewing angles. Multi-view information fusion suppresses dynamic illumination noise. In addition, for persistent fixed turbidity interference, a laser scanning verification mode overcomes the limitations of passive optical imaging on underwater debris detection quality. This invention avoids ineffective verification and energy waste caused by resource misallocation in accordance with operational decisions. Attached Figure Description

[0019] Figure 1 This is a module connection diagram of the computer vision-based waste detection system of the present invention; Figure 2 This is a flowchart illustrating the process of determining the risk disturbance state based on the interference distribution correlation index and the related interference difference index of the detection execution target according to the present invention. Figure 3 This is a flowchart illustrating the process of determining whether to switch from environmental interference analysis to dynamic disturbance analysis when performing environmental interference analysis on a target based on the risk disturbance state of the target. Figure 4 This is a schematic diagram of the computer vision-based garbage detection method of the present invention. Detailed Implementation

[0020] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0021] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0022] It should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0023] Please see Figures 1 to 3 As shown, this embodiment of the invention provides a computer vision-based waste detection system, comprising: The operation detection module is used to periodically determine whether to conduct risk analysis for the waste detection operation process based on the operation interference trend index of the detection execution target; The identification and analysis module, which is connected to the operation detection module, is used to determine whether to switch to dynamic disturbance analysis when performing environmental interference analysis on the target based on the risk disturbance state of the target. The risk disturbance state is determined based on the interference distribution correlation index and the correlation interference difference index of the target. A dynamic disturbance assessment module, which is connected to the identification and analysis module, is used to perform a verification operation based on the interference response assessment parameters of the disturbance space of the relevant interference points corresponding to the detection target. The interference response assessment parameters are determined based on the relevant assessment difference index and the time-series correlation assessment index, or by referring to the relevant difference index and the time-series correlation difference index. An interference disturbance assessment module, which is connected to the identification and analysis module, is used to determine whether to detect the interference fluctuation index of the target for the evaluation execution cycle based on the interference fluctuation correlation parameters, so as to perform a verification operation detection on the distribution analysis space of the target for the evaluation execution cycle.

[0024] This invention is used to monitor and optimize the underwater waste detection process, avoiding the long-term low reliability of waste detection quality caused by fluctuations in image acquisition quality due to spatial differences in water conditions during the detection process. Underwater waste detection devices used to monitor the waste detection process are all designated as detection targets. These targets are all operational equipment capable of detecting and identifying waste in the underwater environment. The invention involves the acquisition of detection execution images from several detection targets. While the invention does not specify the exact structure or model of the detection targets, it is necessary to enable the acquisition and analysis of detection execution images. These detection execution images are images acquired by the detection targets during the process of detecting and identifying waste in the currently detected water area. In this invention, a visual recognition model is used to analyze the images obtained from various detection targets to achieve underwater debris detection. The training process of the visual recognition model includes: data preparation and annotation; using an underwater camera or robot to capture videos / images covering different water qualities (clear / turbid), lighting, and depth scenes; using LabelImg or CVAT tools to draw bounding boxes for debris targets (such as plastic bottles, fishing nets, metal cans, etc.) in each image; exporting the bounding boxes as YOLO format txt annotation files (containing category numbers, normalized center coordinates, and width and height); and then... The data is randomly divided into training, validation, and test sets in a ratio of 8:1:1 or 7:2:1, and organized into the directory structure required by YOLO. Underwater image preprocessing addresses common underwater image problems such as color distortion (blue / green tint), low contrast, and blurriness, requiring targeted enhancement. This involves using the gray-world method or histogram equalization (only for the luminance channel) to restore true colors, applying contrast-limited adaptive histogram equalization (CLAHE) to enhance local details, performing slight sharpening filtering to improve edge clarity, and using the Retinex algorithm to separate reflection components. Data augmentation employs online... Data augmentation libraries (such as Albumentations) are used to randomly apply geometric transformations, color transformations, and special augmentations in each training round. For model selection and configuration, YOLOv8 is selected, and pre-trained models on the COCO dataset are loaded for transfer learning (such as YOLOv5s and YOLOv8n). Hyperparameters are set, the input image resolution is 640×640 pixels, the batch size is set to 8, 16, or 32 depending on the GPU memory, the number of training rounds is 100~300, the optimizer is SGD or AdamW, the learning rate is initially 0.01, and cosine annealing scheduling is used. Model training and validation involve specifying the training and validation set paths and the number of classes in the data configuration YAML file, calling the YOLO training script, setting the hyperparameters and pre-training weight paths, and observing the mAP@0.5, mAP@0.5:0.95, and various losses (bounding box loss, classification loss) on the validation set in real time. Inference is then run on the test set to calculate precision, recall, and mAP. The trained PyTorch model is exported to TorchScript, ONNX, or TensorRT format for cross-platform deployment on NVIDIA Jetson series (underwater robots), Raspberry Pi, or edge computing devices, and inference is performed using OpenCVDNN or ONNX Runtime. These processes are already known to those skilled in the art and will not be elaborated upon here.

[0025] Specifically, the identification and analysis module responds to the analysis execution conditions and performs risk analysis on the waste detection operation process targeting the detection execution target; The analysis execution condition is that the operation interference trend index determined by the evaluation execution cycle for the detection execution target is greater than the preset operation interference trend index. The operation interference trend index is determined based on the execution interference evaluation index determined by each interference evaluation cycle for the detection execution target within the interference evaluation stage.

[0026] In this invention, the detection of garbage images during the operation of detecting the target is periodically evaluated to determine the operation interference trend index of the target. Specifically, an interference evaluation cycle is set, and each interference evaluation cycle detects the operation interference trend index of each target. A value for the duration of the interference evaluation cycle is provided, which is 45s. If the current time is the end time of an interference evaluation cycle, this interference evaluation cycle is recorded as the evaluation execution cycle. For a single detection execution target, the execution interference evaluation index of the detection execution target is determined based on the detection execution images acquired for each detection execution target within the evaluation execution cycle. The operation interference trend index is... , This represents the average execution interference assessment index determined for the detection execution target across all interference assessment cycles except the assessment execution cycle within the interference assessment phase. To evaluate the execution interference evaluation index determined by the execution cycle for the detection execution target, the execution interference evaluation index is the average value of the edge evaluation parameters of the detection execution image obtained by the evaluation execution cycle for the detection execution target. The end time of the interference evaluation phase is the end time of the evaluation execution cycle. The duration of the interference evaluation phase can be set by the user according to actual needs. The higher the user's requirements for the overall detection quality and overall detection efficiency of the underwater garbage detection process, the longer the duration of the interference evaluation phase. A value for the duration of the interference evaluation phase is provided, which is 10 times the duration of the interference evaluation cycle. For a single detection execution image, the detection execution image is converted into a single-channel grayscale image, and convolved with the image using a 3×3 or 5×5 Laplacian kernel (or Gaussian Laplacian LoG) to obtain an edge response map. The edge evaluation parameter is the standard deviation of the pixel values ​​of each pixel in the edge response map. The value of the preset operation interference trend index can be understood as follows: the higher the user's requirements for the overall detection quality and efficiency of the underwater garbage detection process, the smaller the value of the preset operation interference trend index. The operation interference trend index characterizes the degree of deterioration in the image acquisition quality of the target for garbage detection within a certain time range. A preset operation interference trend index value of 0.7 is provided.

[0027] Specifically, the risk disturbance state includes static risk disturbance state and dynamic risk disturbance state; The detection execution target under the static risk disturbance state is the detection execution target where the interference distribution correlation index is less than or equal to the preset interference distribution correlation index and the correlation interference difference index is less than or equal to the preset correlation interference difference index; The detection execution target under the dynamic risk disturbance state is the detection execution target where the interference distribution correlation index is greater than the preset interference distribution correlation index or the correlation interference difference index is greater than the preset correlation interference difference index.

[0028] In this context, for a single detection target, the interference distribution correlation index is the number of detection targets existing in the distribution analysis space besides the target itself during the interference evaluation phase. Detection targets existing in the distribution analysis space are denoted as interference-related targets. The distribution analysis space is the set of point perturbation spaces for each image acquisition point within the evaluation execution cycle. The image acquisition point is the location of the detection target during image acquisition. For a single image acquisition point, the point perturbation space is a spherical space centered at the image acquisition point with the perturbation interference distance as the radius of the sphere. The value of the perturbation interference distance can be set by the user according to actual needs and operational conditions. The higher the user's requirements for the overall detection quality and efficiency of the underwater garbage detection process, the larger the value of the perturbation interference distance. One perturbation interference distance value is provided: 2m. The correlation interference difference index is the average value of the interference difference indices of the interference-related targets existing with the target. For a single interference-related target, the interference difference index... , The interferometry assessment index is determined for the interferometry assessment period corresponding to the existence of the interferometry-related target in the distribution analysis space. To evaluate the execution interference assessment index determined by the execution cycle for this detection execution objective; The values ​​of the preset interference distribution correlation index and the preset interference difference index are understood to reflect the user's higher requirements for the overall detection quality and efficiency of the underwater waste detection process. The higher the user's requirements for the overall detection quality and efficiency, the lower the value of the preset interference distribution correlation index and the preset interference difference index. The interference distribution correlation index characterizes the risk level of the area affected by external operations in the image acquisition process during the evaluation execution cycle. The interference difference index characterizes the degree of influence suffered by the area affected by the image acquisition process during the evaluation execution cycle. One preset interference distribution correlation index value is 4, and another preset interference difference index value is 0.35.

[0029] Specifically, the dynamic disturbance assessment module performs dynamic disturbance analysis on the detection execution target that is in the dynamic risk disturbance state; The setting method for the interference response evaluation parameters of each point disturbance space of the detection execution target is determined based on the interference-related dispersion parameters and interference-related overlap parameters of the detection execution target.

[0030] Specifically, for a single detection target under dynamic risk disturbance, since the interference distribution correlation index of the detection target is greater than the preset interference distribution correlation index or the correlation interference difference index is greater than the preset correlation interference difference index, the area involved in the image acquisition task during the evaluation execution cycle is at greater risk of being affected by the operation of other detection targets, or the area involved in the image acquisition task during the evaluation execution cycle is greatly affected. Therefore, dynamic disturbance analysis is performed on the detection target to evaluate the sensitivity and recovery ability of image acquisition conditions in different water areas to external operation interference. In addition, this avoids the waste of analysis resources caused by performing environmental interference analysis on all detection targets under analysis execution conditions. For a single detection target, the interference correlation dispersion parameters , The target of this detection is the distribution interval index between relevant interference points in the evaluation execution cycle. The interference correlation overlap parameter is used to evaluate the distribution interval index between image acquisition points in the execution cycle for the target of this detection. , The number of interference-related targets present in the target of this detection. To determine the number of relevant interference points for the detection execution target in the evaluation execution cycle, for a single interference-related target, the image acquisition point corresponding to the minimum value of the shortest interval distance between the image acquisition points existing in the distribution analysis space of the detection execution target and the image acquisition points existing in the evaluation execution cycle of the detection execution target is recorded as the relevant interference point. For any number of image acquisition points, all pairs of different acquisition point combinations are traversed, and the shortest interval distance between the two points in each combination is calculated. From all the calculated pairwise point distances, the shortest interval distance with the largest value is selected and recorded as the distribution interval index.

[0031] Specifically, the dynamic disturbance assessment module determines the interference response assessment parameters of the point disturbance space corresponding to the relevant interference points where the dispersed radiation target exists, based on the relevant assessment difference index and the time-series correlation assessment index. The dynamic disturbance assessment module determines the interference response assessment parameters of the disturbance space corresponding to the relevant interference points where the overlapping radiation target exists, based on the reference correlation difference index of each relevant interference point. Based on the temporal correlation difference index, the interference response evaluation parameters of the point disturbance space corresponding to the relevant interference points with temporal correlation difference indices that are less than or equal to the preset temporal correlation difference index for overlapping radiation targets are reduced and adjusted.

[0032] Wherein, the dispersed radiation target is a detection execution target whose interference-related dispersion parameter is greater than a preset interference-related dispersion parameter and whose interference-related overlap parameter is less than or equal to a preset interference-related overlap parameter; the overlapping radiation target is a detection execution target whose interference-related dispersion parameter is less than or equal to a preset interference-related dispersion parameter or whose interference-related overlap parameter is greater than a preset interference-related overlap parameter. For a single detection execution target under dynamic risk disturbance, if the interference correlation dispersion parameter of the detection execution target is greater than the preset interference correlation dispersion parameter and the interference correlation overlap parameter is less than or equal to the preset interference correlation overlap parameter, it indicates that the distribution dispersion of the relevant interference points in the evaluation execution cycle of the detection execution target is high, and there is no high degree of overlap interference of the interference-related targets corresponding to the relevant interference points. That is, the operation of the interference-related targets of the detection execution target has a relatively dispersed impact on the image acquisition quality of the waste detection process. For the point disturbance space corresponding to a single relevant interference point of the detection execution target, the ratio of the correlation evaluation difference index determined for the point disturbance space corresponding to the relevant interference point to the time-series correlation evaluation index is normalized. The value obtained after normalization is recorded as the interference response evaluation parameter of the point disturbance space corresponding to the relevant interference point. The time-series correlation evaluation index... , To determine the minimum time interval between the last moment when each interference-related target exists in the perturbation space corresponding to the interference point and the last moment when image acquisition is performed at the interference point, this is the determination of the relevant interference point location. The duration of the interference assessment period, and the corresponding assessment difference index for the relevant interference point. , The average value of the edge evaluation parameters of the detection execution image obtained at the relevant interference point for the detection target. To evaluate the minimum value of the edge evaluation parameters of the detected execution image obtained during the execution cycle; For a single detection execution target under dynamic risk disturbance, if the interference correlation dispersion parameter of the detection execution target is less than or equal to the preset interference correlation dispersion parameter or the interference correlation overlap parameter is greater than the preset interference correlation overlap parameter, it indicates that the distribution dispersion of the relevant interference points of the detection execution target in the evaluation execution cycle is low, or that the degree of overlap interference of the interference-related targets corresponding to the relevant interference points is high. That is, the operation of the interference-related targets of the detection execution target has a relatively concentrated impact on the image acquisition quality of the waste detection process, and there is interference superposition. For the point disturbance space corresponding to a single relevant interference point of the detection execution target, the reference correlation difference index determined for the point disturbance space corresponding to the relevant interference point is normalized. The value obtained after normalization is recorded as the interference response evaluation parameter of the point disturbance space corresponding to the relevant interference point. The reference correlation difference index is the average value of the evaluation difference index of each interference-related target of the detection execution target when the relevant interference point is determined. For a single interference-related target, the evaluation difference index... , The average value of the edge evaluation parameters of the detection execution image obtained at the relevant interference point for the detection target. The average value of the edge evaluation parameters of the detection execution image obtained within the point perturbation space of the interferometric correlation target is the temporal correlation difference index. , The average of the temporal correlation indices for each group of overlapping time points is used to determine the last moment in the perturbation space of each interfering target that exists at the corresponding interfering point when the relevant interfering point is determined. This moment is denoted as the interfering operation moment. All pairs of adjacent interfering operation moments are traversed and denoted as overlapping time combinations. The interval between the two interfering operation moments in each overlapping time combination is calculated and denoted as the temporal correlation index of the corresponding overlapping time combination. The duration of the interference assessment period; in this invention, the normalization process adopts maximum-minimum normalization, which maps the values ​​to the [0,1] interval.

[0033] For a single overlapping radiation target, if the temporal correlation difference index of the relevant interference points of the overlapping radiation target is less than or equal to the preset temporal correlation difference index, it indicates that the concentration of interference superposition from other detection execution targets at the relevant interference points is high. Therefore, the interference response evaluation parameters of the disturbance space corresponding to the relevant interference points are reduced. The adjusted interference response evaluation parameters are... , The parameters are: unadjusted interference response evaluation parameters and adjustable control parameters. Users can set the value of these adjustable control parameters according to their actual needs. The higher the user's requirements for the overall detection quality and efficiency of the underwater waste detection process, the smaller the value of the adjustable control parameter. One possible value for the adjustable control parameter is 2. The preset time-series correlation difference index is understood to be lower as the user's requirements for the overall detection quality and efficiency of the underwater waste detection process increase. The time-series correlation difference index characterizes the concentration of interference superposition from other detection targets at the relevant interference points. One possible value for the preset time-series correlation difference index is 0.05. The values ​​of the preset interference-related dispersion parameter and the preset interference-related overlap parameter are understood to reflect the user's higher requirements for the overall detection quality and efficiency of the underwater waste detection process. The higher the user's requirements for the overall detection quality and efficiency, the smaller the values ​​of the preset interference-related dispersion parameter and the preset interference-related overlap parameter. The interference-related dispersion parameter characterizes the degree of dispersion of the detection target within the evaluation cycle, while the interference-related overlap parameter characterizes the degree of overlap interference between the corresponding interference targets. One preset interference-related dispersion parameter value is 0.3, and another preset interference-related overlap parameter value is 0.15.

[0034] Specifically, the dynamic disturbance assessment module performs a verification operation based on the verification interval parameter to detect the disturbance space of relevant interference points existing in the target detection execution; The verification interval parameter of any point disturbance space is positively correlated with the corresponding interference response evaluation parameter of the point disturbance space.

[0035] Specifically, for the point disturbance space of a single related interference point of the detection execution target in the dynamic risk disturbance state, if the duration of any detection execution target in the point disturbance space is longer than the review interval parameter, a review operation is performed on the point disturbance space, that is, the detection execution image is re-acquired for the point disturbance space. The review interval parameter is set for the point disturbance space. , These are the parameters for evaluating the interference response determined in the space of the disturbance at this point. The benchmark verification interval parameter is set by the user according to the actual situation. The higher the user's requirements for the overall detection quality and efficiency of the underwater waste detection process, the larger the value of the benchmark verification interval parameter. One benchmark verification interval parameter value is provided, which is 8 minutes.

[0036] Specifically, the interference disturbance assessment module performs environmental interference analysis on the detection execution target that is in the static risk disturbance state; The interference floating correlation parameters are determined based on the interference floating parameters of each image acquisition point.

[0037] The determination of whether to detect the interference float index of the target for evaluation of the execution cycle distribution analysis space is based on the interference float correlation parameters.

[0038] Specifically, the interference disturbance assessment module detects the interference float index in the distribution analysis space where the interference float correlation parameter is greater than the preset interference float correlation parameter; The interference floating index is determined based on the illumination index floating parameters of each image acquisition point.

[0039] Specifically, for a single detection target in the static risk disturbance state, since the interference distribution correlation index of this detection target is less than or equal to the preset interference distribution correlation index and the correlation interference difference index is less than or equal to the preset correlation interference difference index, the risk of the area involved in the image acquisition task during the evaluation execution cycle being affected by the operations of other detection targets, as well as the degree of influence on the area involved in the image acquisition task during the evaluation execution cycle, are relatively small. The impact on the waste detection quality of this type of detection target is unrelated to the water body impact caused by the operation process of other monitoring targets. Therefore, environmental interference analysis is performed on this detection target to optimize the waste detection operation process in the corresponding water space, ensuring the overall analysis quality and efficiency of waste detection.

[0040] For the distribution analysis space of a single detection execution target in the aforementioned static risk disturbance state, the interference floating correlation parameter , The average value of the interference floating parameter for each image acquisition point in the distribution analysis space of the detection execution target is used. For a single image acquisition point, the determination process of the interference floating parameter includes: acquiring the previously acquired detection execution images in the point perturbation space corresponding to the image acquisition point; arranging the illumination index parameters and edge evaluation parameters of the acquired detection execution images according to the acquisition time sequence; acquiring the parameter sequence data corresponding to the illumination index parameters and edge evaluation parameters; inputting the acquired parameter sequence data corresponding to the illumination index parameters and edge evaluation parameters into the Pearson correlation coefficient formula for calculation; and recording the acquired value as the interference floating parameter of the image acquisition point. For a single detection execution image, the illumination index parameter is the average value of the brightness values ​​of each pixel in the detection execution image. If the interference float correlation parameter is less than or equal to the preset interference float correlation parameter, it indicates that the image quality acquired by the detection target in the corresponding distribution analysis space is highly correlated with the fluctuation of external sunlight. Therefore, the interference float index of the distribution analysis space of the detection target is detected. The interference float index is the average value of the illumination index float parameter of each image acquisition point in the distribution analysis space of the detection target. For a single image acquisition point, the illumination index float parameter... , This represents the standard deviation among the illumination parameters of each previously acquired detection execution image within the perturbation space corresponding to the image acquisition point. This is the average value of the illumination index parameters of each detection execution image previously acquired within the disturbance space corresponding to the image acquisition point.

[0041] The value of the preset interference floating correlation parameter can be understood as follows: the higher the user's requirements for the overall detection quality and efficiency of the underwater garbage detection process, the smaller the value of the preset interference floating correlation parameter. The interference floating correlation parameter represents the degree of correlation between the acquired image quality and the fluctuation of external sunlight. A preset value of 0.3 is provided.

[0042] Specifically, the interference disturbance assessment module performs a verification operation based on a stereo vision verification method to detect the distribution analysis space of the detection execution target where the interference fluctuation index is greater than the preset interference fluctuation index; The verification perspective parameters are determined based on the interference floating correlation parameters of the distribution analysis space corresponding to the detection execution target.

[0043] Specifically, for a single detection target, if the interference fluctuation index of the detection target is greater than the preset interference fluctuation index, it indicates that the image acquisition task in the distribution analysis space corresponding to the detection target is subject to a high degree of sporadic interference. This suggests that the probability of the interference affecting the image acquisition task in the distribution analysis space corresponding to the detection target being mitigated over time is relatively high. However, because it is susceptible to interference from changes in external light, when performing a verification operation on the distribution analysis space corresponding to the detection target, a stereo vision verification method is used. This involves acquiring multiple viewpoints of the detection execution images to perform the verification operation on the distribution analysis space corresponding to the detection target. The verification viewpoint parameter is the number of viewpoints corresponding to the detection execution images acquired during the verification operation. The verification viewpoint parameter is negatively correlated with the interference fluctuation correlation parameter. For the verification viewpoint parameter of the distribution analysis space corresponding to the detection target... , The benchmark composite viewpoint parameter is used as the reference. The value of the benchmark composite viewpoint parameter can be set by the user according to the actual situation. One value of the benchmark composite viewpoint parameter is provided, and the value of the benchmark composite viewpoint parameter is 2. For a single detection target, if the interference fluctuation index of the detection target is less than or equal to the preset interference fluctuation index, it indicates that the interference experienced by the image acquisition task in the distributed analysis space corresponding to the detection target is not accidental. This suggests that the probability of the interference level in the image acquisition task in the distributed analysis space corresponding to the detection target being smoothed out over time is low, implying that the interference source is not accidental light fluctuations, but more likely persistent physical factors such as water turbidity and suspended matter. Since laser scanning is unaffected by light changes, can penetrate water with a certain turbidity, and accurately acquires three-dimensional contours, laser scanning is used to verify the distributed analysis space corresponding to the detection target, ensuring the contour recognition quality of the required garbage items. The preset interference fluctuation index is understood to be determined by the user's higher requirements for the overall detection quality and efficiency of the underwater garbage detection process; the higher the value of the preset interference fluctuation index, the greater the value. The interference fluctuation index characterizes the degree of accidental interference experienced by the image acquisition task corresponding to the distributed analysis space. A preset interference fluctuation index value of 0.35 is provided.

[0044] Please see Figure 4 As shown, this is a schematic diagram of the computer vision-based garbage detection method of the present invention. The present invention also provides a computer vision-based garbage detection method, including: Periodically determine whether to conduct risk analysis for the waste detection process based on the operational interference trend index of the detection target; When conducting risk identification analysis, the risk disturbance state is determined based on the interference distribution correlation index and the relevant interference difference index of the detection target, and the environmental interference analysis for the detection target is changed to dynamic disturbance analysis based on the risk disturbance state of the detection target. When performing dynamic disturbance analysis, the interference response evaluation parameters of the disturbance space of the relevant interference points corresponding to the detection target are determined based on the relevant evaluation difference index and the time-series correlation evaluation index, or by referring to the relevant difference index and the time-series correlation difference index, so as to perform the verification operation detection. When performing environmental interference analysis, the interference float correlation parameters are used to determine whether to detect the interference float index of the target for evaluation of the distribution analysis space of the execution cycle, so as to perform a verification operation for the target for evaluation of the distribution analysis space of the execution cycle.

[0045] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A computer vision-based waste detection system, characterized in that, include: The operation detection module is used to periodically determine whether to conduct risk analysis for the waste detection operation process based on the operation interference trend index of the detection execution target; The operation interference trend index is determined based on the execution interference evaluation index of the detection execution target for each interference evaluation cycle within the interference evaluation phase. If the current time is the end time of an interference evaluation cycle, the interference evaluation cycle is recorded as the evaluation execution cycle. For a single detection execution target, the execution interference evaluation index of the detection execution target is determined based on the detection execution images acquired by the detection execution target each time within the evaluation execution cycle. The execution interference evaluation index is the average value of the edge evaluation parameters of the detection execution target for the detection execution image acquired in the evaluation execution cycle. The edge evaluation parameters are the standard deviation of the pixel values ​​of each pixel in the edge response map obtained by convolving the detection execution image with a Laplacian kernel or a Gaussian Laplacian kernel after converting it into a single-channel grayscale image. The identification and analysis module, connected to the job detection module, is used to determine whether to switch to dynamic disturbance analysis when performing environmental interference analysis on the target based on the risk disturbance state of the target. The risk disturbance state is determined based on the interference distribution correlation index and the correlation interference difference index of the target. The interference distribution correlation index is the number of targets existing in the distribution analysis space other than the target during the interference assessment phase. Targets existing in the distribution analysis space are denoted as interference-related targets. The distribution analysis space is the set of point disturbance spaces of each image acquisition point during the assessment execution cycle. The correlation interference difference index is the average value of the interference difference indices of the interference-related targets present in the target. The interference difference index is determined based on the difference between the execution interference assessment index determined for the target during the corresponding interference assessment cycle and the execution interference assessment index determined for the target during the assessment execution cycle. A dynamic disturbance assessment module, connected to the identification and analysis module, performs dynamic disturbance analysis on the detection execution target in a dynamic risk disturbance state. This analysis is used to perform a verification operation based on the interference response assessment parameters of the disturbance space corresponding to the relevant interference points of the detection execution target. The interference response assessment parameters are determined based on a relevant assessment difference index and a time-series correlation assessment index, or by referring to the relevant difference index and the time-series correlation difference index. The relevant assessment difference index is the correlation index of the difference between the average value of the edge assessment parameters of the detection execution image acquired at the relevant interference point and the minimum value of the edge assessment parameters of the detection execution image acquired during the assessment execution cycle. The time-series correlation assessment index is the time interval between the last moment when all interference-related targets of the detection execution target exist in the disturbance space corresponding to the relevant interference point and the last moment when image acquisition is performed at the relevant interference point. The minimum interval duration is the ratio of the duration of the interferometric evaluation cycle; the reference correlation difference index is the average of the evaluation difference indices of all interfering related targets existing in the detection execution target when the relevant interferometric point is determined; the evaluation difference index is the correlation index of the difference between the average edge evaluation parameters of the detection execution image obtained by the detection execution target at the relevant interferometric point and the average edge evaluation parameters of the detection execution image obtained by the interfering related target in the point perturbation space of the relevant interferometric point; the temporal correlation difference index is the ratio of the average interval duration of each adjacent moment in the point perturbation space corresponding to the relevant interferometric point when the relevant interferometric point is determined to the duration of the interferometric evaluation cycle; the relevant interferometric point refers to the image acquisition point that is spatially closest to any interfering related target among multiple image acquisition points of the detection execution target within the evaluation execution cycle; An interference disturbance assessment module, connected to the identification and analysis module, performs environmental interference analysis on the detection execution target in a static risk disturbance state. It determines, based on interference float correlation parameters, whether to detect the interference float index of the detection execution target relative to the distribution analysis space of the assessment execution cycle, and performs a review operation detection on the distribution analysis space of the detection execution target. The interference float correlation parameter is the absolute value of the difference between the average value of the interference float parameters of each image acquisition point in the distribution analysis space of the detection execution target and 1. The interference float parameter is determined based on the Pearson correlation coefficient between the illumination index parameters and edge evaluation parameters of the detection execution images previously acquired within the disturbance space corresponding to the image acquisition point. The illumination index parameter is the average brightness value of each pixel in the detection execution image. The interference float index is the average value of the illumination index float parameters of each image acquisition point in the distribution analysis space of the detection execution target. The illumination index float parameter is the ratio of the standard deviation of the illumination index parameters of each detection execution image previously acquired within the disturbance space corresponding to the image acquisition point to the average value of the illumination index parameters.

2. The computer vision-based waste detection system according to claim 1, characterized in that, The identification and analysis module responds to the analysis execution conditions and performs risk analysis on the waste detection operation process targeting the detection execution target; The analysis execution condition is that the operation interference trend index determined by the evaluation execution cycle for the detection execution target is greater than the preset operation interference trend index. The operation interference trend index is determined based on the execution interference evaluation index determined by each interference evaluation cycle for the detection execution target within the interference evaluation stage.

3. The computer vision-based waste detection system according to claim 2, characterized in that, The risk disturbance state includes static risk disturbance state and dynamic risk disturbance state; The detection execution target under the static risk disturbance state is the detection execution target where the interference distribution correlation index is less than or equal to the preset interference distribution correlation index and the correlation interference difference index is less than or equal to the preset correlation interference difference index; The detection execution target under the dynamic risk disturbance state is the detection execution target where the interference distribution correlation index is greater than the preset interference distribution correlation index or the correlation interference difference index is greater than the preset correlation interference difference index.

4. The computer vision-based waste detection system according to claim 3, characterized in that, The dynamic disturbance assessment module performs dynamic disturbance analysis on the detection execution target that is in the dynamic risk disturbance state; The setting method for the interference response evaluation parameters of each point disturbance space of the detection execution target is determined based on the interference-related dispersion parameters and interference-related overlap parameters; Wherein, the interference correlation dispersion parameter is the ratio of the distribution interval index between the relevant interference points of the detection execution target with respect to the evaluation execution cycle to the distribution interval index between the image acquisition points of the detection execution target with respect to the evaluation execution cycle; the distribution interval index is obtained by traversing all pairwise different acquisition point combinations, calculating the shortest interval distance between the two points in each combination, and selecting the shortest interval distance with the largest value from all the calculated pairwise point distances; the interference correlation overlap parameter is the ratio of the difference between the number of interference-related targets of the detection execution target and the number of relevant interference points of the detection execution target with respect to the evaluation execution cycle, divided by the number of interference-related targets of the detection execution target.

5. The computer vision-based waste detection system according to claim 4, characterized in that, The dynamic disturbance assessment module determines the interference response assessment parameters of the disturbance space of the relevant interference points corresponding to the existence of the dispersed radiation target based on the relevant assessment difference index and the time-series correlation assessment index; wherein, the dispersed radiation target is a detection execution target whose interference-related dispersion parameter is greater than the preset interference-related dispersion parameter and whose interference-related overlap parameter is less than or equal to the preset interference-related overlap parameter; The dynamic disturbance assessment module determines the interference response assessment parameters of the disturbance space corresponding to the relevant interference points where the overlapping radiation target exists based on the reference correlation difference index of each relevant interference point, and determines whether to adjust the interference response assessment parameters of the disturbance space corresponding to each relevant interference point based on the time-series correlation difference index; wherein, the overlapping radiation target is a detection execution target whose interference correlation dispersion parameter is less than or equal to the preset interference correlation dispersion parameter or whose interference correlation overlap parameter is greater than the preset interference correlation overlap parameter; Based on the temporal correlation difference index, the interference response evaluation parameters of the point disturbance space corresponding to the relevant interference points with temporal correlation difference indices that are less than or equal to the preset temporal correlation difference index for overlapping radiation targets are reduced and adjusted.

6. The computer vision-based waste detection system according to claim 5, characterized in that, The dynamic disturbance assessment module performs a verification operation based on the verification interval parameter to detect the disturbance space of the relevant interference points corresponding to the detection target. The verification interval parameter of any point disturbance space is positively correlated with the corresponding interference response evaluation parameter of the point disturbance space; The verification interval parameter is determined based on the benchmark verification interval parameter and the interference response evaluation parameter determined for the disturbance space at that point.

7. The computer vision-based waste detection system according to claim 6, characterized in that, The interference disturbance assessment module performs environmental interference analysis on the detection execution target that is in the static risk disturbance state; The interference floating correlation parameters are determined based on the interference floating parameters of each image acquisition point.

8. The computer vision-based waste detection system according to claim 7, characterized in that, The interference disturbance assessment module detects the interference float index in the distribution analysis space where the interference float correlation parameter is greater than the preset interference float correlation parameter; The interference floating index is determined based on the illumination index floating parameters of each image acquisition point.

9. The computer vision-based waste detection system according to claim 8, characterized in that, The interference disturbance assessment module performs a verification operation based on a stereo vision verification method to detect the distribution analysis space of the detection execution target where the interference float index is greater than the preset interference float index. The verification perspective parameters are determined based on the interference floating correlation parameters of the distribution analysis space corresponding to the detection execution target.

10. A garbage detection method applied to the computer vision-based garbage detection system according to any one of claims 1-9, characterized in that, include: Periodically determine whether to conduct risk analysis for the waste detection process based on the operational interference trend index of the detection target; When conducting risk identification analysis, the risk disturbance state is determined based on the interference distribution correlation index and the relevant interference difference index of the detection target, and the environmental interference analysis for the detection target is changed to dynamic disturbance analysis based on the risk disturbance state of the detection target. When performing dynamic disturbance analysis, the interference response evaluation parameters of the disturbance space of the relevant interference points corresponding to the detection target are determined based on the relevant evaluation difference index and the time-series correlation evaluation index, or by referring to the relevant difference index and the time-series correlation difference index, so as to perform the verification operation detection. When performing environmental interference analysis, the interference float correlation parameters are used to determine whether to detect the interference float index of the target for evaluation of the execution cycle distribution analysis space, so as to perform a verification operation detection for the target for evaluation of the execution cycle distribution analysis space.

Images