Urban sanitation garbage whole-process closed-loop management system based on unmanned aerial vehicle cooperative patrol

Abstract

The application provides an urban environmental sanitation garbage whole-process closed-loop management system based on unmanned aerial vehicle cooperative patrol, relates to the technical field of data processing, and comprises a sampling module, which is used for spatial clustering of geographic position information, forming a garbage accumulation area, and selecting five spatial sampling points according to a preset spatial sampling rule with the center of the garbage accumulation area as a starting position, obtaining diffusion trend data and accumulation density data of the garbage accumulation area by calculating spatial coordinate data variation characteristics of the five sampling points. The application constructs a closed-loop management from garbage data acquisition, identification, accumulation characteristic analysis, state evaluation, clean-up task planning and distribution to effect verification through unmanned aerial vehicle cooperative patrol, can determine garbage disposal priority, efficiently schedule clean-up resources and guarantee clean-up quality, and improves the fine level and operation efficiency of urban environmental sanitation management.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a closed-loop management system for the entire process of urban sanitation waste based on drone-based collaborative patrols. Background Technology

[0002] In the field of urban sanitation management, the application of drones for garbage inspection has been a beneficial attempt. Existing technical solutions typically use drones to capture images of the area and utilize image recognition technology to discover and locate garbage dumps. However, a potential shortcoming of these existing solutions is that they mainly focus on identifying the existence and current location of garbage, while their ability to assess the dynamic development and severity of garbage accumulation is often limited. Specifically, existing technologies struggle to quantify and assess the spread trend and local accumulation density of garbage dumps, which is crucial for determining disposal priorities and rationally allocating collection and transportation resources.

[0003] For example, during an inspection of an old residential area, current technology may simultaneously identify two garbage collection points, A and B. Point A is a relatively large but stable pile of discarded furniture; point B is a smaller pile of household waste that is rapidly spreading into the surrounding alleys. Since current technology mainly relies on visual volume for judgment, it may mark point A as the more important target. However, in reality, the garbage at point B, due to its spread and perishability, has a greater immediate impact on environmental sanitation and residents' travel, and its disposal may be more urgent. This potential judgment bias can affect the scientific allocation of collection tasks, resulting in limited sanitation resources not being invested in the most urgently needed locations. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a closed-loop management system for the entire process of urban sanitation waste based on drone collaborative patrols, so as to avoid patrol omissions caused by scattered waste not being aggregated and improve the integrity of management coverage.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] The first aspect is a closed-loop management system for the entire process of urban sanitation waste based on drone-based collaborative patrols, including:

[0007] The acquisition module is used to acquire initial image data and initial three-dimensional spatial coordinate data within the target area, and to perform image enhancement processing on the image data to obtain the image to be analyzed.

[0008] The identification module is used to identify garbage targets in the image to be analyzed, and to obtain garbage targets, their visual attributes and geographical location information.

[0009] The sampling module is used to spatially cluster geographical location information to form garbage accumulation areas. Starting from the center of the garbage accumulation area, five spatial sampling points are selected according to preset spatial sampling rules. By calculating the spatial coordinate data change characteristics of the five sampling points, the diffusion trend data and accumulation density data of the garbage accumulation area are obtained.

[0010] The evaluation module is used to correlate and integrate the visual attributes of the waste target, the diffusion trend data of the waste accumulation area, and the accumulation density data to evaluate the accumulation status of the waste target.

[0011] The calculation module is used to integrate the accumulation status of waste targets with geographical location information to determine the urgency sequence of disposal for each waste accumulation point and the collection task instruction set, where the assignment operation of the collection task instruction set corresponds to an assignment node.

[0012] The dispatch module is used to assign corresponding cleaning execution units to the cleaning task instruction set, and to perform status tracking and task progress management of the cleaning process.

[0013] The verification module is used to verify the cleaning effect after the cleaning task is completed, and to send the verification information back to the cleaning task instruction set dispatch node to complete the management closed loop.

[0014] In a second aspect, a computing device includes:

[0015] One or more processors;

[0016] A storage device for storing one or more programs that, when executed by one or more processors, enable the one or more processors to implement the system.

[0017] Thirdly, a computer-readable storage medium storing a program that, when executed by a processor, implements the system.

[0018] The above-described solution of the present invention has at least the following beneficial effects:

[0019] By quantifying accumulation density and diffusion trends, combined with visual attributes, a multi-dimensional evaluation matrix is ​​formed. This accurately distinguishes between stable and diffuse accumulation, ensuring that disposal priorities are tilted towards waste sites with greater impact and higher urgency, avoiding judgment bias. After collection, an effectiveness report is generated by comparing initial and verified data and sent back to the dispatch nodes, forming a complete management loop. This not only confirms whether waste has been properly collected but also provides data support for optimizing inspection frequency and scheduling logic. By generating a sequence of disposal urgency, collection execution units with corresponding capabilities are automatically matched according to collection volume and route, and progress is dynamically tracked. This allows for... Limited sanitation resources are directed to high-demand areas to reduce empty runs and inefficient operations. Multiple drones are used for collaborative patrols to collect initial data, which is then aggregated using density clustering algorithms to identify garbage accumulation areas, rather than isolated individual garbage points. This provides more comprehensive coverage of garbage distribution in complex areas, avoiding omissions due to scattered, unaggregated garbage and improving the integrity of management coverage. Progress is tracked through real-time data uploads from collection units and drone-updated patrol data. Any anomalies, such as vehicles deviating from their routes or operations stalling, can be immediately alerted and addressed, reducing collection delays caused by anomalies and ensuring tasks are carried out in an orderly manner according to priority. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of a closed-loop management system for the entire process of urban sanitation waste based on drone collaborative patrol, provided by an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the process of performing garbage target identification processing on the image to be analyzed, and obtaining garbage targets, their visual attributes and geographical location information, according to an embodiment of the present invention. Detailed Implementation

[0022] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0023] like Figure 1 As shown, embodiments of the present invention propose a closed-loop management system for the entire process of urban sanitation waste based on drone collaborative patrols, including:

[0024] The acquisition module is used to acquire initial image data and initial three-dimensional spatial coordinate data within the target area, and to perform image enhancement processing on the image data to obtain the image to be analyzed.

[0025] The identification module is used to identify garbage targets in the image to be analyzed, and to obtain garbage targets, their visual attributes and geographical location information.

[0026] The sampling module is used to spatially cluster geographic location information to form garbage accumulation areas. Starting from the center of the garbage accumulation area, five spatial sampling points are selected according to preset spatial sampling rules. By calculating the spatial coordinate data change characteristics of the five sampling points, the diffusion trend data and accumulation density data of the garbage accumulation area are obtained.

[0027] The evaluation module is used to correlate and integrate the visual attributes of the waste target, the diffusion trend data of the waste accumulation area, and the accumulation density data to evaluate the accumulation status of the waste target.

[0028] The calculation module is used to integrate the accumulation status of waste targets with geographical location information to determine the urgency sequence of disposal for each waste accumulation point and the collection task instruction set, where the assignment operation of the collection task instruction set corresponds to an assignment node.

[0029] The dispatch module is used to assign corresponding cleaning execution units to the cleaning task instruction set, and to perform status tracking and task progress management of the cleaning process.

[0030] The verification module is used to verify the cleaning effect after the cleaning task is completed, and to send the verification information back to the cleaning task instruction set dispatch node to complete the management closed loop.

[0031] In this embodiment of the invention, by quantifying the accumulation density and diffusion trend, and combining visual attributes, a multi-dimensional evaluation matrix is ​​formed. This matrix can distinguish between stable accumulation and diffuse accumulation, ensuring that disposal priority is tilted towards waste points with greater impact and higher urgency, thus avoiding judgment bias. After the waste is cleared, an effect report is generated by comparing the initial data with the verification data and sent back to the dispatch node, forming a complete management closed loop. This not only confirms whether the waste has been cleared properly but also provides data support for optimizing inspection frequency and scheduling logic. By generating a sequence of disposal urgency, the system automatically matches the corresponding capacity of the waste clearing execution unit according to the amount and route cleared, and dynamically tracks the progress. This allows limited sanitation resources to be allocated to high-demand areas, reducing empty runs and inefficient operations. Using multiple drones for collaborative patrols to collect initial data, combined with density clustering algorithms to aggregate garbage accumulation areas, rather than identifying isolated garbage points, provides more comprehensive coverage of garbage distribution in complex areas, avoiding patrol omissions due to scattered, unaggregated garbage and improving the integrity of management coverage. Progress is tracked through real-time data uploads from collection units and drone-updated patrol data, allowing for immediate alerts and handling of any anomalies, such as vehicles deviating from their routes or operations stalling. This reduces collection delays caused by anomalies and ensures tasks are carried out in an orderly manner according to priority.

[0032] In a preferred embodiment of the present invention, initial image data and initial three-dimensional spatial coordinate data within the target area are acquired, and image enhancement processing is performed on the image data to obtain the image to be analyzed, including:

[0033] Multiple drones conduct collaborative patrols of the target area along preset flight paths, collecting initial imagery and 3D spatial coordinate data. Specifically, during drone coordination preparation, 3 to 5 drones equipped with image acquisition and 3D positioning capabilities are selected. Flight paths are preset based on the GIS map and obstacle distribution data of the target area. The path planning must cover the entire target area, using a grid-like cross-coverage pattern. The horizontal flight interval between adjacent drones is calculated by dividing the horizontal width of the target area by the number of drones. The vertical flight height difference is controlled within 10 to 15 meters to avoid flight conflicts and data overlap / redundancy. During the collaborative flight execution phase, a synchronized start command is sent to all drones through the drone ground control center, ensuring that all drones take off simultaneously along the preset paths and maintain time synchronization. During the flight, the ground control center receives real-time positioning data from each UAV. If a UAV deviates from the preset path, a correction command is automatically sent to bring it back to the path, ensuring the integrity of the patrol coverage. In the initial data acquisition phase, each UAV collects initial image data during flight using its onboard high-definition optical camera. The image resolution is set to 1920×1080 pixels or higher, and the frame rate is set to 1 frame every 2 seconds. At the same time, it collects initial three-dimensional spatial coordinate data using its onboard LiDAR module. The acquisition frequency is consistent with the image acquisition frame rate to ensure that each frame of image corresponds to a set of three-dimensional coordinate data. The collected initial image data and initial three-dimensional spatial coordinate data are transmitted back to the ground control center in real time via wireless transmission and stored as structured data files according to the UAV number, acquisition time, and coordinate range.

[0034] The initial image data undergoes illumination equalization and noise filtering to obtain a uniform image for analysis. Specifically, the illumination equalization process employs the Retinex algorithm. This algorithm addresses the uneven illumination issues commonly found in urban sanitation images, such as building shadows and backlit areas. Its core logic is to separate the illuminance component dominated by ambient light from the reflectance component determined by the material properties of the target, thus completely eliminating the interference of light intensity differences on image quality while preserving the detailed features of the garbage target. First, a weighted average method is used to convert the initial color image into a single-channel grayscale image. This is achieved by calculating the grayscale value as: grayscale value = 0.299 × red channel pixel value + 0.587 × green channel pixel value + 0.114 × blue channel pixel value. Subsequently, a natural logarithmic transformation is performed on each pixel of the grayscale image to convert the multiplicative relationship between the original image pixel value and the illuminance component pixel value into an additive relationship in the logarithmic domain, i.e., ln(original pixel value) = ln(illuminance component pixel value) + ln(reflectance component pixel value). This transformation simplifies the separation calculation of the two types of components. Next, a filter kernel is selected based on the local detail scale of the image: a 5×5 Gaussian filter kernel is used for areas with dense garbage texture, and a 7×7 Gaussian filter kernel is used for areas with rich garbage contour details. The logarithmically transformed image is then subjected to neighborhood weighted average smoothing to filter out high-frequency detail information of the garbage target. The logarithmic form of the illuminance component with a gradual change is extracted, and then an inverse transformation of this logarithmic form is performed through exponential operation, i.e., illuminance component pixel value = The initial grayscale image is restored to its original scale using the illuminance component. Then, the corresponding pixel value of the initial grayscale image is divided pixel by pixel by this illuminance component to obtain the reflectance component, which reflects the essential characteristics of the garbage target. All three components maintain the inherent relationship: original image pixel value = illuminance component pixel value × reflectance component pixel value. Subsequently, the effective area of ​​the initial image is traversed pixel by pixel. Invalid pixels within a 10-pixel margin due to shooting distortion are excluded. The grayscale values ​​of all effective pixels are then counted and summed to obtain the total grayscale value. This total grayscale value is divided by the total number of effective pixels to obtain the overall grayscale mean of the initial image. Simultaneously, a target is set based on the human visual comfort range of 120 to 150 and the grayscale clarity requirements for urban sanitation garbage identification. The grayscale mean is matched one by one with the corresponding pixel value of the original illuminance component according to the pixel position. The pixel correction is completed by multiplying the pixel value of the original illuminance component by the ratio of the target grayscale mean to the grayscale mean of the initial image. This ensures that the illuminance component of each pixel in the entire area is adjusted according to a uniform ratio, avoiding local overbrightness that leads to the loss of garbage texture details or underbrightness that leads to the blurring of garbage outlines. Finally, the corrected illuminance component and the original reflection component are multiplied point by point according to the same pixel coordinate position. All pixel values ​​after multiplication are normalized and the pixel values ​​are accurately mapped to the standard grayscale range of 0 to 255. This process is then used to generate an image with uniform illumination distribution and complete preservation of garbage edge texture and color details.

[0035] Noise filtering is performed on the image after illumination equalization using a Gaussian filtering algorithm. This specifically removes Gaussian noise and salt-and-pepper noise generated during image acquisition due to sensor electronic interference and ambient light fluctuations. First, the grayscale variance of the image after illumination equalization is calculated. A variance less than 50 is considered weak noise, and a 3×3 filter kernel is selected. A variance greater than or equal to 50 is considered strong noise, and a 5×5 filter kernel is selected. The Gaussian function expression for the filter kernel is as follows: ,in Represents the coordinates on the filter kernel The corresponding weight value, Pi It is a natural constant. The standard deviation of the Gaussian function is used to control the smoothness of the filter kernel. Its value ranges from 0.8 to 1.2 and increases synchronously with the increase of noise intensity. After calculating the weight values ​​of each position of the filter kernel according to this expression, all weight values ​​are normalized to ensure that the sum of the weights is 1, avoiding image brightness shift. The selected filter kernel covers each pixel of the illumination-equalized image in the order from left to right and from top to bottom, so that the center of the filter kernel is aligned with the current processing pixel. The product of each weight value in the filter kernel and the corresponding image pixel value is calculated, and then all product results are summed. Finally, the sum is divided by the sum of the weights of all elements of the filter kernel to obtain the filtered pixel value of the current pixel. Random noise in the image is completely removed through pixel-by-pixel convolution operation, while preserving the edge details of the garbage targets to the greatest extent. The final output is a uniform quality image with complete details of the garbage targets without noise interference after illumination equalization and noise filtering.

[0036] This embodiment, through multi-drone collaborative patrols combined with preset path intervals and altitude control, avoids blind spots in area coverage during single-drone patrols, solving the problem of incomplete coverage in existing single-patrol methods. Illumination equalization processing addresses the brightness differences in initial images caused by uneven illumination, such as building shadows and backlit areas. Noise filtering removes interference noise from the images. Together, these two processes ensure uniform image quality and clear details, avoiding the low accuracy of garbage target identification due to poor image quality in existing technologies, and providing reliable data support for garbage feature extraction. The output images can be directly input into the identification module for garbage target recognition. The high-quality initial data formed by these two processes ensures the operational efficiency of the entire closed-loop management system.

[0037] like Figure 2 As shown, in another preferred embodiment of the present invention, the image to be analyzed is subjected to garbage target identification processing to obtain garbage targets, visual attributes of garbage targets, and geographical location information, including:

[0038] Multi-scale feature analysis is performed on uniform quality images to extract color, texture, and shape features of targets. Specifically, a multi-scale feature analysis framework is constructed based on the uniform quality images. A Gaussian pyramid layering strategy is used to generate three image layers at different scales to cover garbage targets of different sizes (small-volume scattered garbage and large-volume piled garbage), avoiding missed or false detections at a single scale. The first layer is the original scale of the image to be analyzed. The second layer is obtained by first applying a 5×5 Gaussian filter (standard deviation σ=0.8) to the original image and then downsampling at a sampling interval of 2, i.e., retaining 1 pixel for every 2 pixels. The third layer is obtained by first applying a 5×5 Gaussian filter (standard deviation σ=1.2) to the second layer image and then downsampling at a sampling interval of 2. The three image layers together constitute the basis of multi-scale analysis.

[0039] During color feature extraction, images at each scale are first converted from the RGB color space to the HSV color space (to reduce the interference of illumination changes on color features). The conversion process is calculated using the following formula: For brightness V, the maximum value among the current pixel's RGB three-channel values ​​(R, G, B, ranging from 0 to 255) is taken, and then divided by 255, i.e., V = max(R, G, B) / 255. If the maximum value of the RGB three channels is not 0, (maximum value - minimum value) is taken and divided by the maximum value, i.e., saturation S = [max(R, G, B) - min(R, G, B)] / max(R, G, B) If the maximum value is 0, then S=0; if the maximum value of the three RGB channels is equal to the minimum value, then H=0; if the maximum value is R, then H=60×[(G-B) / (max(R,G,B)-min(R,G,B))]; if the maximum value is G, then hue H=60×[(B-R) / (max(R,G,B)-min(R,G,B))+2]; if the maximum value is B, then H=60×[(R-G) / (max(R,G,B)-min(R,G,B))+4]; if the calculated H<0, then 360 needs to be added to make H fall within the range of 0 to 360.

[0040] After color space conversion, potential target regions in images at various scales are initially located by adaptive threshold segmentation (using the local grayscale mean of the image as the threshold). Histograms of the H, S, and V channels are calculated for each target region. The H channel is divided into 16 intervals, and the S and V channels are each divided into 8 intervals. The mean and variance of the histogram for each channel are calculated. The mean and variance of the H, S, and V channels at the three scales (a total of 3 scales × 3 channels × 2 parameters = 18 parameters) are concatenated to form the color feature vector of the target.

[0041] Texture feature extraction employs the Local Binary Pattern (LBP) algorithm to extract texture features from target regions at various image scales. Specifically, for each pixel within the target region, a 3×3 neighborhood is taken centered on that pixel. Using the grayscale value of the center pixel as a threshold, the grayscale values ​​of the eight neighboring pixels are compared to this threshold. Values ​​greater than or equal to the threshold are recorded as 1, and values ​​less than the threshold are recorded as 0. The eight comparison results are arranged in clockwise or counterclockwise order to form an 8-bit binary number, which is then converted to decimal to obtain the LBP value of the current pixel. The distribution of LBP values ​​across all pixels within the target region is statistically analyzed, generating an LBP histogram containing 256 intervals (corresponding to LBP values ​​from 0 to 255). Based on this histogram, two texture parameters are calculated: energy is obtained by summing the squares of the frequencies across all intervals, and entropy is obtained using the formula entropy. The calculation yielded, where It is the first The frequency of a certain interval, if the frequency of a certain interval If the value is 0, the corresponding term in that interval is counted as 0; the LBP histogram energy and entropy (a total of 3 scales × 2 parameters = 6 parameters) at the 3 scales are concatenated to form the texture feature vector of the target.

[0042] During shape feature extraction, the edges of the target region in the image at various scales are obtained using the Canny edge detection algorithm. The specific steps are as follows: First, a 3×3 Gaussian filter (standard deviation σ=1.0) is used to smooth and denoise the target region, reducing noise interference with edge detection. Then, the horizontal gradient (Gx) and vertical gradient (Gy) of each pixel are calculated using the Sobel operator, and the gradient magnitude is calculated as follows: Calculate the gradient magnitude of each pixel and determine the gradient direction. Perform non-maximum suppression on the gradient magnitude, retaining only pixels with local maximum values ​​in the gradient direction and removing non-edge pixels. Finally, use double threshold segmentation, setting the high threshold to 0.2 × the maximum gradient magnitude of the image and the low threshold to 0.1 × the maximum gradient magnitude of the image. Retain pixels with gradient magnitudes higher than the high threshold and connect pixels with gradient magnitudes between the high and low thresholds that are connected to pixels with high thresholds to obtain the target edge contour.

[0043] After obtaining the target edge contour, four core shape parameters are calculated step by step based on this contour. The specific calculation process is as follows: Area calculation uses a region filling algorithm (such as the seed filling method) to determine the internal region of the edge contour. Taking any pixel within the contour as the seed point, it expands outward to the surrounding 8-neighborhood, marking all pixels surrounded by the contour. Then, the total number of marked pixels is counted, and this number is the area of ​​the region surrounded by the edge contour. This ensures that the statistical range only includes valid pixels inside the contour, excluding the contour line itself and external pixels. Perimeter calculation uses a contour tracing algorithm (such as the 8-neighborhood contour tracing method) to traverse the pixels on the edge contour. That is, starting from the starting pixel of the contour, it traces the adjacent contour pixels in a clockwise direction, recording the coordinates of each pixel and avoiding duplicate counting. The final total number of traced pixels is the perimeter of the edge contour, ensuring complete coverage of all boundary pixels of the contour. Circularity calculation is based on the obtained area and perimeter, using the formula: Circularity = 4 × π × Area ÷ ( The formula for calculating the roundness is used, where π is the mathematical constant for pi, approximately 3.1416. This parameter is dimensionless. If the target is a standard circle, the roundness is close to 1. If the target is elongated or irregular, such as scattered plastic waste, the roundness will be significantly less than 1. This parameter can be used to distinguish the morphological differences between spherical waste and sheet-like or strip-shaped waste. The rectangularity calculation first determines the minimum bounding rectangle through a rotation calibration method. That is, based on the pixel coordinates of all edge contours, the bounding rectangle is calculated under different rotation angles (0° to 90°, step size 1°). The rectangle area is calculated by selecting the smallest rectangle as the minimum bounding rectangle, measuring its length (length of the longer side) and width (length of the shorter side), and then multiplying them. The rectangle degree is calculated as: Rectangularity = Area ÷ (Length of minimum bounding rectangle × Width of minimum bounding rectangle). The closer the rectangularity is to 1, the closer the target outline is to a rectangle, such as a regular building component. If the target is irregular in shape, such as scattered kitchen waste, the rectangularity will be significantly smaller. This parameter can be used to distinguish the difference in the regularity of the shape between garbage and background objects such as buildings, vegetation, etc.

[0044] Among the four parameters, area and perimeter are significantly affected by the target scale. For example, the area and perimeter values ​​of the same piece of garbage differ significantly in images at different scales. Circularity and rectangularity, on the other hand, are dimensionless parameters, which are more robust to scale changes and can more accurately reflect the morphological differences of the garbage target, such as the irregular shape of scattered garbage and the blocky shape of piled garbage, which are clearly distinguished from the regular shape of building components and the natural shape of vegetation in the background. Therefore, circularity and rectangularity are selected as the core shape features. The parameters are stitched together in the order of the original scale, the second scale, and the third scale. That is, the circularity and rectangularity of the original scale are stitched together first, then the circularity and rectangularity of the second scale are stitched together, and finally the circularity and rectangularity of the third scale are stitched together, forming a target shape feature vector with a total of 3 scales × 2 parameters = 6 parameters.

[0045] A waste target recognition model is established based on the color, texture, and shape features of the target. By comparing the color, texture, and shape features of the target with a pre-set waste feature library, scattered waste targets are distinguished from background objects such as vegetation and buildings. The visual attribute information of the identified waste targets and their corresponding geographical coordinates are recorded. Specifically, the model includes: first, constructing a pre-set waste feature library, which involves collecting typical waste samples, including plastic waste, paper waste, metal waste, and kitchen waste, and background object samples, including vegetation, building walls, roads, and facility components. Color feature vectors, texture feature vectors, and shape feature vectors of the two types of samples are extracted respectively. Then, all sample features are stored in a structured format according to sample type, color feature vector, texture feature vector, and shape feature vector to form the waste feature library.

[0046] The model input data is a fused feature formed by concatenating color feature vectors, texture feature vectors, and shape feature vectors in the order of color feature vectors, texture feature vectors, and shape feature vectors. This fused feature is a 30-dimensional vector with clearly defined dimensions. The color feature vector contains the mean and variance of the H, S, and V channels at three scales (3 scales × 3 channels × 2 parameters = 18 dimensions). The texture feature vector contains the energy and entropy of the LBP histogram at three scales (3 scales × 2 parameters = 6 dimensions). The shape feature vector contains the roundness and rectangularity at three scales (3 scales × 2 parameters = 6 dimensions). The concatenation of these three features forms a unified 30-dimensional feature vector, ensuring the integrity and standardization of the model input. Support is constructed based on this 30-dimensional fused feature. The Support Vector Machine (SVM) classification model, also known as the waste target recognition model, is implemented as follows: First, all sample features are normalized, that is, the feature values ​​of each dimension are mapped to the interval [0, 1] to eliminate the influence of scale differences between different feature dimensions on model training. Then, the training sample types are divided. The 30-dimensional fused features of typical waste samples (plastic waste, paper waste, metal waste, kitchen waste, etc.) are set as positive samples, and the 30-dimensional fused features of background object samples (vegetation, building walls, road surfaces, facility components, etc.) are set as negative samples to ensure that the sample categories cover the core objects of the waste recognition scenario. The model uses a radial basis function (RBF kernel) to construct the feature mapping relationship. The kernel function expression is as follows: Calculation, where =1 / feature dimension, that is =1 / 30; , Given the feature vectors of two samples, The Euclidean distance between two vectors is used. The model penalty parameter C is adjusted through 5-fold cross-validation. The samples are first evenly divided into 5 groups. Each time, 4 groups are selected as the training set and 1 group as the validation set, repeating this process 5 times. The C value is tested within a range of 0.01 to 100. This parameter controls the balance between classification error and model complexity. Specifically, a smaller C value results in stronger regularization and a simpler model structure, but is more prone to underfitting. A larger C value results in weaker regularization and a stronger model fit, but is more prone to overfitting. During cross-validation, the garbage target identification on the validation set is used. Accuracy, precision (number of correctly identified junk samples ÷ total number of samples the model classifies as junk), and recall (number of correctly identified junk samples ÷ total number of actual junk samples) are used as comprehensive evaluation indicators. Finally, the C value that achieves the optimal balance among the three indicators (prioritizing recall of no less than 90% and precision of no less than 85%) is selected as the optimal penalty parameter. The optimal C value is substituted into the model, and all positive and negative samples are used for final training until the model converges (the loss value of the training set tends to stabilize, and the change in loss value is less than 1e-5 for 5 consecutive iterations), thus obtaining the trained junk target recognition model.

[0047] The 30-dimensional feature vector (color, texture, and shape spliced ​​vector) of the target extracted from the image to be analyzed is first compared with all feature vectors in the pre-set garbage feature database using a cosine similarity calculation. The similarity calculation formula is: Similarity = ,in Let be the feature vector of the target to be analyzed. The model uses the sample feature vectors from the waste feature library. The previously concatenated 30-dimensional feature vector of the target (containing color, texture, and shape parameters at three scales) is input into the trained waste target recognition model. The radial basis function kernel is used to calculate the mapping relationship between the target features and the training sample features, outputting a binary classification label (waste label or background label). Simultaneously, the model retrieves the cosine similarity between the target's 30-dimensional feature vector and the 30-dimensional feature vectors of all waste samples in the preset waste feature library. The maximum value is extracted as the optimal matching similarity between the target and the waste sample. The judgment logic is as follows: if the model outputs a waste label and the optimal matching similarity is ≥0.7, it indicates that the target highly matches typical waste samples at the feature level, and the target is determined to be scattered waste. If the model outputs a background label, or the optimal matching similarity is <0.7, it indicates that the target features are closer to the feature patterns of background objects such as vegetation, building walls, roads, and facility components, and it is determined to be a background object. Through a dual verification method of model classification and feature similarity, the model distinguishes waste targets from background objects.

[0048] For objects identified as scattered garbage targets, their visual attribute information is fully recorded. This information consists of core parameters after feature fusion at various scales, including the mean and variance of the H channel, S channel, and V channel in color features; the energy and entropy of the LBP histogram in texture features; and the roundness and rectangularity in shape features. Simultaneously, based on the image pixel coordinates (u, v) - 3D spatial coordinates (X, Y, Z) mapping relationship established through camera calibration and 3D reconstruction, the pixel region corresponding to the garbage target in the image is located (based on the center pixel coordinates of the target edge contour). The 3D spatial coordinates corresponding to the pixel position are matched from the initial 3D spatial coordinate data, and these coordinates are officially recorded as the geographical location coordinates of the scattered garbage target.

[0049] This embodiment uses multi-scale feature analysis to cover waste targets of different sizes, avoiding the missed detection of small, scattered waste or the false detection of large, piled-up waste by a single scale. It also integrates color, texture, and shape features to solve problems such as the difficulty of distinguishing similarly colored waste from vegetation or similarly textured waste from building components using a single visual feature, reducing recognition errors caused by background interference. The extracted color, texture, and shape features constitute the visual attributes of the waste targets. Combined with the recorded geographical coordinates, this provides basic data for spatial clustering and multi-dimensional evaluation in the sampling module, avoiding misjudgments of pile status due to missing features in existing technologies. It employs HSV color space and LBP texture algorithm, which are robust to changes in lighting and environmental noise, solving the problem of poor recognition stability in complex environments. It accurately distinguishes waste targets and records their geographical coordinates, avoiding misallocation of collection resources due to target misjudgment or missing coordinates.

[0050] In a preferred embodiment of the present invention, spatial clustering of geographical location information is performed to form a garbage accumulation area. Starting from the center of the garbage accumulation area, five spatial sampling points are selected according to a preset spatial sampling rule. By calculating the spatial coordinate data change characteristics of the five sampling points, the diffusion trend data and accumulation density data of the garbage accumulation area are obtained, including:

[0051] Density clustering (DBSCAN) is used to cluster the geographic coordinates of waste targets, aggregating spatially adjacent waste targets into waste accumulation areas. Specifically, this involves: first, organizing and recording the three-dimensional geographic coordinates (X, Y, Z) of all scattered waste targets to form a waste target coordinate set. The coordinate units are consistent with the initial collected three-dimensional spatial coordinates. After forming the coordinate set, the DBSCAN algorithm is used for clustering analysis based on this set. The core idea is to aggregate adjacent waste targets into waste accumulation areas based on spatial location correlation. The specific implementation process is as follows: first, determine the key clustering parameters. The neighborhood radius ε is determined by calculating the average nearest neighbor distance of all waste target coordinates. The calculation method is to traverse each waste target, find its nearest other waste target's spatial distance (calculated using the three-dimensional Euclidean distance formula), and calculate the mean of all nearest neighbor distances. ε is set as 1.2 times this mean. The minimum number of core points, MinPts, is adaptively adjusted according to the total number of waste targets, calculated as MinPts = 3 + ... This parameter ensures that clustering can effectively distinguish between densely packed areas and isolated, scattered waste.

[0052] After the parameters are determined, a density clustering process is executed. Each waste target in the coordinate set is traversed, and a 3D neighborhood is constructed with that target as the center and ε as the radius. The number of other waste targets contained within the neighborhood is counted. If the number of targets within the neighborhood is greater than or equal to MinPts, then that target is the core point, and all targets within its neighborhood are marked as density reachable points. The process continues to traverse the marked density reachable points. If the number of targets within the neighborhood of a reachable point is greater than or equal to MinPts, then the targets within its neighborhood are also marked as density reachable points. This diffusion process is repeated until no new density reachable points are discovered. At this point, all density-connected targets constitute a waste target set, i.e., a single waste accumulation region. The above clustering operation is repeated for all unmarked waste targets to complete the aggregation of all waste accumulation regions. Finally, the subset of waste target coordinates and a unique region identifier corresponding to each accumulation region are output.

[0053] Based on the waste accumulation area, a convex hull boundary is constructed for the waste accumulation area. Starting from the geometric center of the convex hull boundary, five spatial sampling points are set along the main distribution direction of the convex hull boundary. The first sampling point is located at the geometric center of the convex hull boundary; the second sampling point is located at half the distance from the geometric center to the convex hull boundary; the third sampling point is located on the convex hull boundary; the fourth sampling point is located one meter away from the convex hull boundary on the extension line of the main distribution direction outside the convex hull boundary; and the fifth sampling point is located two meters away from the convex hull boundary on the extension line of the main distribution direction outside the convex hull boundary. Specifically, after completing the aggregation of waste accumulation areas, for each subset of waste target coordinates in the accumulation area, convex hull boundary construction and spatial sampling point setting are carried out. The specific process is as follows: First, convex hull boundary construction is performed. Two-dimensional coordinates of the horizontal plane (X, Y axes) are extracted from the three-dimensional coordinate subset of the accumulation area (ignoring the Z-axis height and focusing on the plane distribution range). The Graham scan method is used to construct the two-dimensional convex hull boundary. The steps are as follows: First, select the point with the smallest Y-axis value in the two-dimensional coordinate set as the starting point. If there are multiple points with the same minimum Y-axis value, select the point with the smallest X-axis value. Sort the remaining points in ascending order according to the polar angle of the line connecting them to the starting point. The polar angle is calculated using arctan2(yi-y0,xi-x0), where (x0,y0) is the coordinate of the starting point and (xi,yi) is the coordinate of the other points. Initialize the convex hull vertex stack and add the sorted points to the stack in sequence. After each addition, check whether the top three points of the stack form a counterclockwise direction (by using the cross product of vectors; if (x2-x1)(y3-y1)-(y2-y1)(x3-x1)>0, then it is counterclockwise). If it is clockwise, remove the middle point. Repeat this process until all points are processed. The vertices in the stack are the two-dimensional vertex coordinates of the convex hull boundary. After obtaining the two-dimensional vertex coordinates, combine them with the mean Z-axis coordinates of all garbage targets in the accumulation area to supplement the three-dimensional information, forming a complete three-dimensional convex hull boundary. The three-dimensional vertex coordinates are (X convex hull, Y convex hull, Z mean).

[0054] Next, calculate the geometric center of the convex hull boundary. The geometric center is represented by three-dimensional coordinates, calculated as follows: Geometric center X coordinate = sum of X coordinates of all convex hull vertices ÷ number of convex hull vertices; Geometric center Y coordinate = sum of Y coordinates of all convex hull vertices ÷ number of convex hull vertices; Geometric center Z coordinate = sum of Z coordinates of all convex hull vertices ÷ number of convex hull vertices, denoted as (Xc, Yc, Zc). Then, determine the principal distribution direction of the convex hull boundary. First, calculate the mean of the two-dimensional coordinates (X, Y) of all vertices of the convex hull boundary (mean X = sum of X coordinates of all vertices ÷ number of convex hull vertices, mean Y = sum of Y coordinates of all vertices ÷ number of convex hull vertices). Then, construct the covariance matrix. The matrix elements are calculated as follows: covariance XX = Sum ÷ (Number of convex hull vertices) 1) Covariance Y = Sum ÷ (Number of convex hull vertices) 1) Covariance XY = (X coordinates of all vertices) (2D coordinate mean X) × (Y coordinates of all vertices) The sum of the mean Y-axis of the two-dimensional coordinates ÷ (number of vertices of the convex hull) 1) Solve for the eigenvalues ​​and eigenvectors of the covariance matrix. The eigenvector corresponding to the largest eigenvalue is the principal distribution direction of the convex hull boundary. Normalize this eigenvector: unit vector = eigenvector ÷ L2 norm of the eigenvector. L2 norm = ... The unit vector of the main distribution direction (dx, dy) is obtained. Finally, spatial sampling points are set. The first sampling point is directly taken as the geometric center (Xc, Yc, Zc) of the convex hull boundary, i.e., (X1, Y1, Z1) = (Xc, Yc, Zc). The second sampling point starts from the geometric center (Xc, Yc) along the unit vector of the main distribution direction (dx, dy), and finds the vertex (Xf, Yf) farthest from the geometric center in the two-dimensional vertices of the convex hull boundary. This distance D is calculated using the two-dimensional Euclidean distance formula: D = The second sampling point has coordinates (Xc+dx×D / 2, Yc+dy×D / 2, Zc), which is located at half the distance from the geometric center to the convex hull boundary. The third sampling point is taken as the three-dimensional coordinates (Xf, Yf, Zc) corresponding to the farthest vertex on the convex hull boundary, i.e., (X3, Y3, Z3) = (Xf, Yf, Zc). The fourth sampling point is set outside the convex hull boundary along the extension of the unit vector (dx, dy) in the main distribution direction, with coordinates (Xf+dx×1, Yf+dy×1, Zc), which is 1 meter away from the convex hull boundary. The fifth sampling point is set outside the convex hull boundary along the extension of the unit vector (dx, dy) in the main distribution direction, with coordinates (Xf+dx×2, Yf+dy×2, Zc), which is 2 meters away from the convex hull boundary.

[0055] The stock density data of the waste accumulation area is obtained by calculating the spatial coordinate data distribution density of the first, second, and third sampling points. Specifically, based on the set first, second, and third sampling points and the determined density clustering neighborhood radius ε, the stock density data of the waste accumulation area is calculated. The specific process is as follows: First, the sampling neighborhood radius is determined by directly using the neighborhood radius ε of the density clustering to ensure that the spatial scale of the density calculation is consistent with the clustering scale, thus improving data reliability. Next, the spatial distribution density of each sampling point is calculated. That is, for the first, second, and third sampling points, a three-dimensional spherical neighborhood is constructed with each sampling point as the center and ε as the radius. The volume of the sphere is calculated using the formula: spherical volume = Calculate and count the number of garbage targets contained in each spherical neighborhood of the garbage accumulation area, denoted as N1 (first sampling point), N2 (second sampling point), and N3 (third sampling point). The accumulation density of each sampling point is calculated as density = number of targets in the neighborhood ÷ spherical volume, i.e., ρ1 = N1 / spherical volume, ρ2 = N2 / spherical volume, and ρ3 = N3 / spherical volume. Finally, the accumulation density data is determined by taking the arithmetic mean of the accumulation densities of the three sampling points as the final accumulation density data of the garbage accumulation area, i.e., accumulation density ρ = (ρ1 + ρ2 + ρ3) / 3. This value can reflect the average level of density distribution from the core to the boundary of the accumulation area, objectively reflecting the degree of garbage concentration.

[0056] By analyzing the spatial coordinate data variation patterns of the first, second, third, fourth, and fifth sampling points, the diffusion trend data of the waste accumulation area is obtained. Specifically, to obtain the diffusion trend data of the waste accumulation area, based on the five sampling points and the determined distance D, and combined with the accumulation density calculation method, the density-distance variation curve is fitted. The specific process is as follows: First, the distance parameters corresponding to each sampling point are calculated. Taking the geometric center (Xc, Yc) of the convex hull boundary as the reference point, the horizontal distance from the five sampling points to the reference point is calculated, where L1=0 (the first sampling point is the reference point), L2=D / 2 (the second sampling point), L3=D (the third sampling point), L4=D+1 (the fourth sampling point), and L5=D+2 (the fifth sampling point), where D is the distance from the determined geometric center to the main distribution direction of the convex hull boundary. Then, the accumulation density of the fourth and fifth sampling points is calculated. That is, a three-dimensional spherical neighborhood is constructed with the fourth and fifth sampling points as the center and ε as the radius, respectively. The number of waste targets N4 and N5 in the accumulation area within the neighborhood is counted, and the corresponding accumulation density is calculated. The density is calculated as ρ4 = N4 / spherical volume and ρ5 = N5 / spherical volume. Next, a density-distance variation curve is fitted, with the distance parameters L (L1 to L5) of the five sampling points as the abscissa and the corresponding packing density ρ (ρ1 to ρ5) as the ordinate. A least squares method is used to fit a linear regression curve, with the regression equation being ρ = a × L + b, where a is the linear regression coefficient (slope) and b is the intercept. The core calculation logic of the least squares method is to minimize the sum of the squared differences between the actual density at each sampling point and the predicted density on the regression curve, thus solving for the minimum... The coefficients a and b are optimized to ensure that the curve accurately reflects the density variation with distance. Finally, the diffusion trend data is determined, that is, the regression coefficient a represents the diffusion trend data of the waste accumulation area. The sign of a indicates the diffusion direction. When a is negative, it means that the density decreases with increasing distance along the main distribution direction, which is consistent with the natural diffusion law of waste. When a is positive, it means that the density increases with increasing distance along the main distribution direction, which is an abnormal accumulation situation. The absolute value of a reflects the diffusion intensity. The larger the absolute value, the more significant the change in density with distance and the more obvious the spatial spread trend of waste.

[0057] This embodiment uses a density clustering algorithm to aggregate waste targets based on spatial location correlation, effectively distinguishing between scattered waste and concentrated accumulations, avoiding misclassification of spatially separated waste as the same accumulation area. The construction of convex hull boundaries accurately delineates the spatial distribution range of waste accumulation areas, and the setting of five sampling points covers the core, middle, boundary, and outer extension areas of the accumulation, ensuring that the density and diffusion trend calculations can comprehensively reflect the internal structure and external extension characteristics of the accumulation area. The accumulation density data quantifies the degree of waste concentration, and the diffusion trend data clarifies the spatial spread characteristics of waste. These two indicators provide a quantitative basis for the objective assessment of the waste accumulation status, avoiding assessment bias caused by subjective judgment. Based on the quantified accumulation density and diffusion trend, resources can be prioritized to treat high-density, high-diffusion-risk accumulation areas, optimize the priority ranking of collection and transportation, improve collection and transportation efficiency, reduce resource waste, and provide data support for the refined management of urban sanitation waste collection and transportation.

[0058] In a preferred embodiment of the present invention, the visual attributes of the waste target, the diffusion trend data of the waste accumulation area, and the accumulation density data are correlated and fused to evaluate the accumulation state of the waste target, including:

[0059] Based on the color distribution features, texture complexity features, and contour features of the visual attributes of waste targets, the diffusion rate and diffusion range in the diffusion trend data of waste accumulation areas, and the density per unit area in the accumulation density data, a multi-dimensional evaluation matrix is ​​constructed. Specifically, this includes: defining the core dimensions and indicators of the multi-dimensional evaluation matrix; all indicators are derived from the processing results of previous steps. Specifically, the focus is first on the visual attribute dimension, which mainly describes the morphology and appearance of the waste itself, containing three core features, all calculated based on the recorded visual attribute information of the waste targets. The color distribution feature is taken as the average of the variances of the H channel, S channel, and V channel in the color features. The magnitude of this average value... The value of small values ​​directly reflects the uniformity of garbage color. The larger the average value, the more significant the color difference of the garbage. For example, the value will be significantly higher for garbage mixed with multiple colors such as plastic, paper, and kitchen waste. The texture complexity feature directly uses the LBP histogram entropy in texture features. The size of the entropy value corresponds to the complexity of the garbage texture. The larger the entropy value, the more complex the garbage texture. For example, the mesh texture of woven bags and the smooth texture of paper can be clearly distinguished by this indicator. The outline feature uses the rectangularity in shape features to measure the regularity of the garbage outline. The smaller the rectangularity, the more irregular the garbage outline. For example, the rectangularity of scattered scraps of garbage and torn pieces of paper is much lower than that of regular cardboard boxes and plastic boxes.

[0060] Next, we address the diffusion trend dimension, which focuses on reflecting the spatial spread of waste. This dimension includes two core indicators, both calculated based on diffusion trend data. The diffusion rate is taken as the absolute value of the linear regression coefficient. This absolute value is positively correlated with the speed of waste diffusion; the larger the absolute value, the more significant the change in waste density with distance, and the faster the diffusion speed. For example, lightweight waste piled up in windy areas will have a higher value. The diffusion range is represented by the horizontal projected area of ​​the convex hull boundary. The calculation uses the two-dimensional coordinates of the convex hull vertices according to the shoelace formula. Specifically, the X-coordinate of each vertex is multiplied by the Y-coordinate of the next vertex. Then subtract the X-coordinate of the next vertex multiplied by the Y-coordinate of the current vertex, add all such calculation results together, take the absolute value, and finally multiply by one-half. The resulting value is the horizontal projected area (where the next vertex after the last vertex is considered the first vertex). The larger the projected area, the wider the range of waste spread. Finally, there is the packing density dimension, which directly reflects the degree of waste concentration. The core indicator is the density per unit area. The calculated packing density can be used directly because it has been standardized by the spherical neighborhood volume and can accurately reflect the amount of waste per unit space without additional adjustment.

[0061] Since the units of the above indicators are different, such as density in cubic meters and area in square meters, direct use in comprehensive calculations will result in deviations due to the difference in dimensions. Therefore, it is necessary to normalize all indicators to eliminate the influence of units and ensure that all indicators are within the same numerical range. After normalizing all indicators, six standardized indicators are obtained, corresponding to the normalized results of color distribution characteristics, texture complexity characteristics, contour characteristics, diffusion rate, diffusion range, and density per unit area, respectively. Next, a multi-dimensional evaluation matrix is ​​constructed. Each row of the matrix represents a waste accumulation area (each area is assigned a unique identifier in sequential order), each column represents a standardized indicator, and each element in the matrix is ​​the specific value of the corresponding waste accumulation area on the corresponding standardized indicator. If the total number of waste accumulation areas is m, the matrix will contain m rows and 6 columns, fully presenting the multi-dimensional characteristics of each area.

[0062] Based on the multi-dimensional evaluation matrix and according to the preset stacking state level classification parameters, the waste stacking state is divided into three levels: light stacking state, moderate stacking state, and heavy stacking state, thus obtaining the stacking state of the waste target. Specifically, based on the constructed multi-dimensional evaluation matrix, through preset classification parameters and weighted calculation, the abstract index data is transformed into specific stacking state levels. The specific process is as follows: First, the core parameters for stacking state level classification are preset. The parameter settings fully combine the actual needs of sanitation operations. Considering that stacking density has the greatest impact on the priority of collection, followed by diffusion-related indicators, weights are assigned to the six standardized indicators (the sum of all weights is 1). Specifically, the weights are: density per unit area 0.3, diffusion rate 0.2, diffusion range 0.15, color distribution feature 0.1, texture complexity feature 0.1, and contour feature 0.15. At the same time, the correspondence between the comprehensive score range and the stacking state is preset, that is, a comprehensive score ≥ 0.7 corresponds to heavy stacking, and 0.4 ≤ comprehensive score ≥ 0.7 corresponds to heavy stacking. A score < 0.7 corresponds to moderate accumulation, and a comprehensive score < 0.4 corresponds to light accumulation. Then, the comprehensive score for each waste accumulation area is calculated using a weighted summation method. This involves multiplying the value of each standardized indicator by its corresponding weight, and then summing all the products. The sum is the comprehensive score for that area, i.e., the overall assessment value. Finally, the accumulation status is classified according to the comprehensive score. When the comprehensive score ≥ 0.7, it is classified as heavy accumulation. This type of accumulation typically has high density per unit area, spreads rapidly, and covers a wide area, such as large garbage piles in residential corners, which have a significant environmental impact and require priority disposal. When 0.4 ≤ comprehensive score < 0.7, it is classified as moderate accumulation. This type of accumulation has moderate density and spread, such as small to medium-sized garbage piles on the roadside, which can be disposed of according to conventional priority. When the comprehensive score < 0.4, it is classified as light accumulation. This type of accumulation has low density and insignificant spread, such as small amounts of scattered plastic bottles and waste paper, which can be disposed of later. The final result is the accumulation status level for each waste accumulation area.

[0063] This embodiment integrates multiple indicators such as visual attributes, diffusion trends, and accumulation density to avoid assessment bias caused by a single indicator and accurately depict the complex state of garbage accumulation. The preset weights and scoring intervals are set based on the actual needs of sanitation operations to ensure standardized classification, reduce the subjectivity of manual judgment, and make the accumulation status of different areas comparable. The clear light, moderate, and heavy levels are directly related to the subsequent urgency ranking logic to ensure that high-risk accumulation areas can be identified first and avoid resource misallocation.

[0064] In a preferred embodiment of the present invention, the accumulation status of the waste target is fused with geographical location information to determine the urgency sequence of each waste accumulation point and the collection task instruction set, wherein the assignment operation of the collection task instruction set corresponds to an assignment node, including:

[0065] Based on the accumulation status of waste targets, a multi-factor evaluation matrix is ​​constructed. Waste accumulation points are ranked from highest to lowest urgency to generate a urgency sequence. Specifically, after obtaining the accumulation status level of each waste accumulation area, a multi-factor evaluation matrix is ​​further constructed, integrating three key factors: accumulation status, geographical location, and surrounding environment. This generates a urgency sequence more closely aligned with actual sanitation operations. The specific process is as follows: First, a multi-factor evaluation matrix is ​​constructed, supplementing the accumulation status with two additional key influencing factors. The definitions and calculation methods for each factor are as follows: The accumulation status factor quantifies the accumulation status level: heavy accumulation is assigned a value of 1.0, moderate accumulation 0.5, and light accumulation 0.2. This quantified value directly reflects the urgent disposal needs of the accumulation itself and is the core of the urgency assessment. The geographical location factor calculates the straight-line distance from the geometric center of the waste accumulation area (a determined geometric center) to the nearest collection station. After normalizing this distance, 1 is subtracted from the normalized distance. The closer the distance, the larger the result of subtracting the normalized distance from 1, representing higher collection efficiency and a higher urgency level. The urgency score can be appropriately increased. The surrounding environment factor is determined based on the background image to determine if the area surrounding the accumulation zone is a densely populated area, such as near residential entrances, schools, or markets. If it is a densely populated area, a value of 1.0 is assigned; otherwise, a value of 0.5 is assigned. This setting prioritizes the disposal of garbage in densely populated areas, reducing the impact on the public environment and residents' health. Weights are then assigned to the three categories of factors (the sum of all weights is 1). The weighting follows the priority logic of sanitation operations: the accumulation status factor has a weight of 0.6, serving as a core influencing factor that directly determines the basic urgency of the accumulation; the geographical location factor has a weight of 0.2 (distance affects the efficiency of waste collection resources; nearby areas can be prioritized); and the surrounding environment factor has a weight of 0.2 (environmental safety is an important supplement; densely populated areas require special attention). Each factor's value is multiplied by its corresponding weight, and all products are summed to obtain the urgency score for that area. After calculating the urgency scores for all accumulation areas, all garbage accumulation areas are sorted from highest to lowest score to generate a disposal urgency sequence. Areas at the top of the sequence are those requiring priority disposal.

[0066] Based on the urgency sequence, a collection task instruction set is automatically generated, containing collection priority parameters, estimated collection volume data, and collection route planning data. This instruction set is then sent to the task assignment node. Specifically, based on the urgency sequence, core operational parameters are further integrated to generate a standardized collection task instruction set, which is then sent to the task assignment node. This provides direct guidance for actual sanitation operations. The specific process is as follows: First, collection priority parameters are generated. These parameters are based on the urgency ranking, using ranking, area identifiers, and geographic coordinates as the core structure to construct an intuitive and unambiguous parameter system, clearly defining the operational priority and target location. The specific ranking of urgency, such as 1st, 2nd, and 3rd, is used as the core priority identifier. The higher the ranking, the higher the collection priority, allowing operators to quickly determine which area to handle first and then... The system determines the execution order for each area. For each ranking, two key pieces of information are associated: first, a unique identifier for the corresponding waste accumulation area, a unique number assigned to each area to effectively prevent confusion caused by proximity or similar appearance; and second, the three-dimensional geographic coordinates of the accumulation area, derived from recorded waste target geographic location data, specifically the three-dimensional coordinates of the geometric center (Xc, Yc, Zc) of the accumulation area. This ensures that operators can directly locate the area on the electronic map using these coordinates, resolving the problem of mislocation due to ambiguous area names or missing landmarks. For example, the accumulation area ranked first in urgency can be clearly described as having a collection priority of level 1; a unique area identifier of 001; and geographic coordinates (X: 123.45, Y: 67.89, Z: 10.11), clearly presenting the priority and directly supporting accurate positioning.

[0067] To avoid a mismatch between the waste collection equipment and the actual amount of waste—such as small-tonnage vehicles being unable to carry heavily piled-up waste while large-tonnage vehicles are used to transport small amounts of waste, resulting in resource waste—a two-step precise estimation of the collection volume is conducted based on previously acquired spatial data of the waste accumulation area. This provides a clear basis for equipment selection. The first step involves basic data collection and confirmation, consisting of two fundamental data points, both derived from previously processed results. The horizontal projection area of ​​the convex hull is calculated using a shoelace formula, which fully covers the planar distribution of the waste accumulation area and accurately reflects the lateral extent of the accumulation. The average height of the waste is calculated by extracting the Z-coordinates (vertical data from the recorded three-dimensional coordinates) of all waste targets within the accumulation area. The specific calculation... The logic is to subtract the minimum value from the maximum value of the Z-coordinate of all garbage targets in the area. For example, if the maximum Z-coordinate of garbage in a certain accumulation area is 12.5 meters and the minimum is 10.2 meters, its average height is 2.3 meters. This value can truly reflect the thickness of the garbage accumulation in the vertical direction. Secondly, the volume calculation and accuracy guarantee adopt the calculation logic of accumulation volume = horizontal projection area of ​​convex hull × average height of garbage. This formula is based on the actual scenario assumption that the garbage accumulation shape is approximately a column. It simplifies the calculation process and can meet the accuracy requirements of sanitation operations for estimating the amount of garbage collected. For example, if the horizontal projection area of ​​the convex hull in a certain area is 15 square meters and the average height is 2 meters, then the garbage accumulation volume in that area is 15 × 2 = 30 cubic meters.

[0068] The process for estimating the weight of collected waste involves combining long-term experience data accumulated from sanitation operations and pre-setting differentiated average densities based on waste type. This avoids weight estimation errors caused by a single density. For mixed waste, which contains multiple components such as plastics, paper, kitchen waste, and metals, the average density is set at 0.5 tons per cubic meter (due to its complex composition and medium density). The average density for pure plastic waste is set at 0.3 tons per cubic meter (due to its lightweight material, high porosity, and relatively low density). The average density for pure kitchen waste is set at 0.8 tons per cubic meter (due to its high water content, compact texture, and relatively high density). These density settings perfectly reflect the physical characteristics of different types of waste. Weight calculation and equipment adaptation are also considered. It is recommended to calculate the transport weight as: transport weight = waste accumulation volume × average density of the corresponding waste type. For example, the transport weight of 30 cubic meters of mixed waste is 30 × 0.5 = 15 tons. Based on the calculation results, equipment matching suggestions are given simultaneously. That is, it is recommended to dispatch a 20-ton transport vehicle for 15 tons of waste, and a small-tonnage tricycle for light accumulation of less than 5 cubic meters. This allows operators to quickly match suitable equipment without additional calculations. The final output transport volume estimation data is uniformly integrated into a standardized format of waste accumulation volume (in cubic meters), transport weight (in tons), and recommended equipment type, which directly serves the equipment scheduling process.

[0069] To maximize collection efficiency and reduce unnecessary mileage and time costs, the operation route is planned using the nearest collection station as the starting point, strictly following the order of urgency. The Dijkstra shortest path algorithm is employed to generate route data that closely reflects actual road conditions. Specifically, the starting point is determined by first extracting the three-dimensional coordinates of all pre-set collection stations in the sanitation operation management system, then calculating the straight-line distance between each station and the geometric center of each garbage accumulation area, and selecting the collection station closest to the current planned area as the route starting point. For example, if the straight-line distance from collection station 1 to area A is 2 kilometers and the straight-line distance from collection station 2 is 5 kilometers, then station 1 is determined as the starting point of the collection route for that area, reducing the initial travel distance from the source.

[0070] The road network topology map construction involves setting the geometric centers of the determined starting point (nearest collection station) and all garbage accumulation areas in the urgency sequence as nodes for route planning. The edge weights between nodes are based on the actual road distance (not the straight-line distance). This distance is obtained by calling professional map service APIs such as Baidu Maps and Gaode Maps, which fully reflects the urban road layout and actual traffic conditions. For example, the straight-line distance between node A (collection station 1) and node B (area 001) is 2 kilometers, but due to factors such as road restrictions and river obstructions, the actual driving distance is 3 kilometers. In this case, the edge weight of this segment is directly set to 3 kilometers to ensure that the route planning does not deviate from the actual traffic scenario.

[0071] Using Dijkstra's algorithm, starting from the starting point, the algorithm strictly plans the path according to the order of urgency. First, it plans the shortest path to priority 1 areas, then the shortest path to priority 2 areas, and so on. For example, if the urgency sequence is area 001 (priority 1), area 003 (priority 2), and area 005 (priority 3), the algorithm will first calculate the shortest path from collection point 1 to area 001, then use area 001 as an intermediate node to calculate the shortest path to area 003, and finally use area 003 as an intermediate node to calculate the shortest path to area 005. This ensures that the path order is completely consistent with the priority, while minimizing the total travel distance and avoiding unnecessary mileage due to disordered path order. The final generated collection route planning data contains three core parts: the first is the path node order, which specifies the exact order of travel, for example, the travel order is collection point 1, then to area 001 (…). X: 123.45, Y: 67.89), then proceed to area 003 (X: 123.56, Y: 67.98), and finally to area 005 (X: 123.67, Y: 68.09); secondly, the actual distance between nodes, that is, the actual road distance between each two adjacent nodes is marked, for example, from waste collection station 1 to area 001, 3 kilometers, from area 001 to area 003, 2.5 kilometers, from area 003 to area 005, 4 kilometers; thirdly, the estimated travel time, that is, based on the traffic scenarios of different road sections, the normal driving speed is preset (25 km / h for urban roads and 40 km / h for suburban roads), and the travel time of each route is estimated by the formula travel time = distance between nodes ÷ corresponding road section driving speed. For example, for a 3-kilometer urban road section, the estimated travel time is 3 ÷ 25 = 0.12 hours, or 7.2 minutes. At the same time, the total travel time is summarized to facilitate the operators to reasonably plan the work schedule for the day.

[0072] The previously generated collection priority parameters, collection volume estimation data, and collection route planning data are integrated into a standardized instruction set according to the fixed format of area identifier, collection priority, collection volume data, and collection route. This ensures that task assignment nodes, such as the sanitation operation management platform, can be directly parsed without additional secondary processing. For example, a complete instruction set entry would be: area unique identifier 001; collection priority level 1; collection volume data is 30 cubic meters in volume, 15 tons in weight, and a recommended 20-ton collection vehicle; the collection route starts at collection station 1 (X: 123.00, Y: 67.00), first proceeding to area 001 (X: 123.45, Y: 67.89), a distance of 3 kilometers, with an estimated travel time of 7.2 minutes; then proceeding to area 003 (X: 123.56, Y: 67.98), a distance of 2.5 kilometers, with an estimated travel time of 6 minutes; through... Pre-defined standardized data interfaces, such as HTTP-based API interfaces, send the integrated waste collection task instruction set to the task assignment node. To ensure the security of core data, SSL encryption technology is used during interface transmission to prevent the leakage of critical information such as geographical coordinates and waste collection weight. After receiving the instruction set, the task assignment node automatically associates it with the personnel and vehicle dispatch information to generate an execution task sheet that can be directly printed or viewed on the operation terminal, realizing a closed loop of the entire process of parameter generation, instruction integration, secure transmission, and task implementation. After completing the calculation of waste collection priority parameters, waste collection volume estimation data, and waste collection route planning data, all information is integrated according to the structured format of area identification, priority, waste collection volume, and route to generate a standardized waste collection task instruction set. This instruction set is then sent to the task assignment node through a pre-defined data interface, such as the API interface of the sanitation operation management platform.

[0073] This embodiment prioritizes urgency based on multiple factors, allowing sanitation resources to be deployed to high-priority areas, improving resource utilization efficiency and reducing the problem of prioritizing the cleanup of lightly accumulated debris in suburban areas while neglecting heavily accumulated debris in urban areas. The estimated sanitation volume data helps workers prepare suitable transportation equipment in advance, such as using small-tonnage vehicles to clean up lightly accumulated debris and large-tonnage vehicles to clean up heavily accumulated debris. Route planning data reduces unnecessary travel and shortens operation time. The structured task instruction set makes task assignment, execution, and supervision traceable, realizing a shift from passive cleanup to proactive planning and promoting the upgrading of sanitation operations towards refinement and intelligence.

[0074] In a preferred embodiment of the present invention, a corresponding cleaning execution unit is assigned to the cleaning task instruction set, and the cleaning process is tracked for status and task progress is managed, including:

[0075] Based on the estimated transport volume and route planning in the transport task instruction set, transport execution units with corresponding transport capabilities and ranges are matched. Specifically, this involves using the estimated transport volume and route planning data from the transport task instruction set as the core basis, and through dual adaptation filtering, matching transport execution units with corresponding transport capabilities and ranges. The specific implementation process is as follows: First, the basic information of each transport execution unit is fully entered into the management system in advance, clarifying two core attributes: transport capability (including maximum carrying volume, maximum carrying weight, and compatible equipment type) and transport range (including the geographical boundary coordinates of the responsible area and the operating radius threshold, which is uniformly set to 20 kilometers to adapt to the coverage area of ​​conventional sanitation operations); based on the estimated transport volume... According to the criteria, the selection of waste collection and transportation units (WPCs) that meet three conditions are as follows: first, the maximum carrying capacity is not less than the estimated waste accumulation volume; second, the maximum carrying weight is not less than the estimated waste collection weight; and third, the type of compatible equipment is completely consistent with the recommended compatible equipment in the waste collection volume estimation. This ensures that the WPC can successfully complete its tasks from a load perspective. The selection of WPCs for suitable transportation range involves first calculating the straight-line distance from the starting point (nearest WPC station) in the WPC route plan to the farthest area to be collected. Then, WPCs that meet two conditions are selected: first, the operating radius threshold (20 kilometers) is not less than this straight-line distance; and second, the geographical coordinates of all areas to be collected are within the boundary coordinates of the geographical area it is responsible for. This ensures that the WPC can reach all operating areas from a spatial coverage perspective.

[0076] The waste collection task instructions are assigned to the matched waste collection execution units, and a correspondence between waste collection tasks and waste collection execution units is established. Specifically, after matching waste collection execution units, task assignment is immediately initiated and a unique association is established. The specific process is as follows: through the scheduling system of the task assignment node, the standardized waste collection task instruction set is pushed to the operating terminal of the successfully matched waste collection execution unit. The terminal includes two types: a vehicle-mounted intelligent terminal (installed in the driver's cab of the waste collection vehicle, allowing the driver to view route planning and coordinates of the area to be cleaned in real time) and a dedicated mobile APP for operators (allowing on-site operators to record waste collection progress and upload operation photos). The instruction set includes area identifiers, cleaning... The system prioritizes transportation, such as Level 1 Priority, Level 2 Regular, and includes transportation volume data (volume, weight, recommended equipment), route planning details (travel sequence from origin to each area, estimated time), and operational requirements, such as sealed transportation of kitchen waste. This ensures that the execution unit obtains comprehensive and accurate operational data. At the same time, the management system automatically generates transportation task and execution unit association data records, which include the unique identifier of the transportation task, the transportation execution unit number, the task assignment time, and the terminal reception status (received / not received / failed to receive; if not received or failed to receive, the system will automatically resend 3 times). This record clarifies the unique correspondence between each transportation task and its corresponding execution unit.

[0077] By combining real-time location and operational data uploaded by the waste removal execution unit with patrol data updated by the drone, the waste removal process is tracked in real time, yielding real-time tracking results, specifically including:

[0078] The system continuously receives two types of core data: first, data uploaded by the waste removal execution unit, namely real-time location data (obtained by the GPS / BeiDou positioning module on the execution unit, with meter-level accuracy in three-dimensional coordinates, updated every 30 seconds to ensure real-time location) and operation data (including vehicle speed, operation start time, area markers for completed waste removal, and current operation status (awaiting departure / driving / operating / paused). Workers manually mark the start and completion of waste removal via a mobile app, and vehicle speed is automatically collected by the vehicle-mounted terminal); second, drone patrol data, where drones patrol and photograph all areas to be removed at a preset frequency (once every 30 minutes), uploading the latest image data (1080P resolution, consistent with the initial image resolution) of areas not yet removed / in progress with waste removal, and re-collected three-dimensional spatial coordinate data (to verify whether the area boundaries have changed), supplementing and verifying the authenticity of the execution unit's operation status.

[0079] The management system synchronously integrates and verifies the received multi-source data. This involves comparing the real-time location of the waste collection unit with the planned route node sequence one by one. For example, if the planned route starts from station A, goes to area 1, and then to area 2, and the unit's location deviates from the coordinates of area 1 by more than 500 meters, it is determined that the unit may have deviated from the route. The system also matches the operational data with preset task progress nodes (arrival at the target area, start of waste collection, completion of waste collection). If the time the unit marks its arrival at area 1 deviates from the planned time by more than 15 minutes, a progress warning is triggered. Finally, the system cross-verifies the data using drone patrol data. If the unit marks its completion of waste collection in area 1, but the drone imagery shows that there is still waste remaining in area 1, the operational status is determined. To prevent discrepancies between reported and actual data and avoid misjudgments caused by a single data source, a comprehensive real-time status tracking result is generated after integration and analysis. The result covers the current operation stage, such as being en route to Area 2, currently clearing in Area 3, the number of areas cleared (e.g., 2 out of 3 areas to be cleared have been completed), task progress (e.g., 67%), whether the route has deviated from the planned route (e.g., no deviation, 300 meters from the coordinates of Area 1), whether the operation has started on time (e.g., on time, 20 minutes delay in clearing in Area 2), and whether the vehicle speed is normal (e.g., 25 km / h in urban areas, meeting the speed limit). All results are presented in real-time in chart form on the management interface of the task assignment node, allowing managers to intuitively grasp the dynamics of the clearing process.

[0080] Based on real-time status tracking results, the progress of the waste removal task is dynamically updated, and warnings and handling of abnormal situations are provided. Specifically, the progress is calculated using the number of areas that have been cleared in the real-time status tracking results as the core data, according to the formula: Task completion progress = Number of areas cleared ÷ Total number of areas to be cleared × 100%. For example, if 2 out of 3 areas to be cleared have been completed, the progress is 2 ÷ 3 × 100% ≈ 67%. The calculation results are synchronized in real time to the management interface of the task assignment node, presented as a percentage and a progress bar, which can be viewed by managers in real time. The progress ranking of all tasks provides a clear understanding of the overall progress; three types of anomaly judgment conditions are preset in advance, covering all operational scenarios: first, route deviation anomaly, that is, the real-time location of the cleaning execution unit deviates from the planned route by more than 500 meters (a 500-meter fault tolerance threshold is set to take into account the need for detours on urban roads); second, time overrun anomaly, that is, the operation time of a single area exceeds the expected time by 30%, such as the expected cleaning time of 40 minutes for area 1, but the actual time exceeds 52 minutes; and third, operation quality anomaly, that is, the drone inspection finds that the cleaning is not carried out as required, such as garbage not being cleaned up or leakage during transportation.

[0081] When any abnormal condition is triggered, the management system immediately initiates a closed-loop process. The first step is automatic early warning, which is displayed in a pop-up window on the management interface. For example, if the execution unit of Task-002 deviates from the route in area 2 by 450 meters, an SMS / APP notification is sent to the management personnel and the person in charge of the execution unit, including the abnormality type, current location, and handling requirements. The second step is to generate targeted handling instructions. For example, if the route deviates, the instruction is to verify the cause of the deviation (such as road construction), adjust to the planned route within 15 minutes, and if adjustment is not possible, feedback is given to the dispatch system for replanning. If the task times out, the instruction is to coordinate the backup execution unit to provide support and prioritize the clearing of the remaining area. The third step is to record the entire process, archiving the time of occurrence of the abnormality, the triggering cause, the handling measures, and the handling time in the management system. For example, if the route is adjusted within 10 minutes, the handling result is recorded. If the route has been returned to the planned route and the subsequent progress has not been affected, a complete abnormality handling closed loop is formed to prevent the abnormality from escalating and causing task delays.

[0082] This embodiment ensures that the cleaning capacity matches the task requirements by matching the cleaning execution units, avoiding resource waste or operation delays caused by insufficient capacity; clear task assignment and corresponding relationship establishment make the operation responsibility clear and traceable, reducing management difficulty; real-time status tracking realizes dynamic monitoring of the cleaning process, ensuring that the operation proceeds in an orderly manner according to plan; the abnormal early warning and handling mechanism effectively avoids execution risks and improves the reliability and overall efficiency of the cleaning operation.

[0083] In a preferred embodiment of the present invention, after the waste collection task is completed, the waste collection effect is verified, and the verification information is sent back to the waste collection task instruction set dispatch node to complete the management closed loop, including:

[0084] After the waste removal task progress data indicates that the task has been completed, the verification data collection for the cleared areas is initiated to obtain the verification image data and verification 3D spatial coordinate data. Specifically, when the task progress data shows that the task completion progress = 100%, that is, after all areas of the task have been cleared, the verification data collection for the cleared areas is immediately initiated to ensure that the collected data is comparable to the initial data in terms of range and accuracy. The specific process is as follows: using the boundary coordinates of the initial waste accumulation area, such as the boundary coordinates of area 1 X1-X5, Y1-Y5, as the benchmark, the boundary line of the verification collection range is generated in the management system. Coordinate matching is used to ensure that the verification range completely coincides with the initial data collection range. For example, if the initial range of area 1 is X: 123.45 to 123.50, Y: 67.89 to 67.94, the verification range remains consistent to avoid distortion of the comparison results due to range deviation (such as over-collection or missed collection areas).

[0085] Two types of core data are collected using drones. First, the drones are used to verify the image data after the cleanup. They fly at the same altitude as the initial image (e.g., 50 meters) and along the same route (e.g., clockwise) to capture full coverage of the cleaned area. The resolution is set to 1080P (consistent with the initial image) to ensure that the image clarity and perspective match the initial data, allowing for direct pixel-level comparison. Second, the drones are used to verify the three-dimensional spatial coordinate data. They are equipped with a lidar module (with centimeter-level accuracy, e.g., ±2 cm) to re-collect the three-dimensional coordinates of the area. This includes the three-dimensional coordinates of the area's geometric center (for verification against the initial center coordinates), ground elevation data (for calculating the ground recovery height after cleanup), and the three-dimensional coordinates of key points, such as the highest and lowest points of the initial garbage accumulation, to verify the thoroughness of the cleanup. All collected data is uploaded to the management system in real time and automatically tagged with verification data labels.

[0086] The post-clearance verification image data is compared and analyzed with the initial image data. The post-clearance verification 3D spatial coordinate data is also compared and analyzed with the initial 3D spatial coordinate data to generate a verification report on the clearance effect. Specifically, this includes: importing the collected post-clearance verification image data and initial image data into an image comparison system for frame-by-frame pixel-level comparison. This involves image feature analysis (combining color differences and contour shapes to distinguish residual garbage from the background; for example, residual garbage often appears as irregular color blocks, while the background is often a uniform road surface or vegetation tone) to automatically identify and extract the pixel regions corresponding to residual garbage in the verification images, and counting the total number of pixels in that region, i.e., the pixel area of ​​the residual garbage. The percentage of residual waste area is then calculated using the formula: Residual waste pixel area ÷ Initial waste accumulation pixel area × 100%. For example, if the initial waste corresponds to a pixel area of ​​10,000 pixels in the image, and the residual waste pixel area in the post-cleanup review image is 300 pixels, then the residual waste area percentage is 3%. At the same time, the type of residual waste is further determined by combining texture and color characteristics. For example, plastic fragments are mostly smooth and reflective textures, light-colored or colored; paper is mostly matte textures, white or light-colored; and kitchen waste is mostly dark-colored and irregularly fibrous textures. The specific distribution location is also marked, such as residual plastic waste in the northeast corner of area 1, forming a complete image comparison result.

[0087] When comparing three-dimensional spatial coordinate data, two sets of key height data are calculated separately: the initial average height of the garbage is calculated based on the recorded initial three-dimensional spatial coordinates, taking the difference between the maximum and minimum initial Z-coordinate values. For example, if the maximum Z-coordinate of the initial area 1 is 12.5 meters and the minimum is 10.2 meters, then the initial average height of the garbage is 12.5 meters - 10.2 meters = 2.3 meters. The verified average height of the ground is calculated based on the verified three-dimensional spatial coordinates collected in step 700, taking the difference between the maximum and minimum verified Z-coordinate values. For example, if the maximum Z-coordinate of the verified area 1 is 1... The minimum height is 0.3 meters, and the maximum is 10.2 meters. Therefore, the average height of the verified ground is 10.3 meters - 10.2 meters = 0.1 meters. The height difference is then calculated using the formula: Height Difference = Initial Average Garbage Height - Verified Average Ground Height. For example, 2.3 meters - 0.1 meters = 2.2 meters. This difference directly reflects the degree of ground restoration after removal. The larger the difference, the closer the ground height is to the original ground elevation, indicating more thorough garbage removal. Simultaneously, the deviation between the three-dimensional coordinates (X-verified, Y-verified, Z-verified) of the geometric center of the verified area and the coordinates (X-initial, Y-initial, Z-initial) of the initial area's geometric center is calculated. The deviation value is determined according to... If the deviation is ≤1 meter (1 meter is set as the coordinate matching fault tolerance threshold), it is determined that the two sets of coordinates correspond to the same area, so as to avoid the accuracy of the comparison results being affected by data confusion.

[0088] The report integrates image comparison results, including the percentage of residual waste area, type of residual waste, specific distribution location, and 3D coordinate comparison results, including height difference and geometric center coordinate deviation. Combined with preset removal effectiveness standards (residual waste area percentage < 5% and height difference < 0.3 meters indicates complete removal; residual waste area percentage 5% to 15% and height difference 0.3 to 0.8 meters indicates partial residue; residual waste area percentage > 15% or height difference > 0.8 meters indicates failure to meet standards), the removal effectiveness level of the current area is determined. The report also needs to clearly indicate the data source, such as the initial image 20250101 corresponding to area 1, and the verification image 202501... 01 Corresponds to Area 1 (after cleanup); Initial 3D spatial coordinates; Verification of 3D spatial coordinates; Explanation of the comparison process, such as image comparison through frame-by-frame pixel-level comparison, combined with color and texture features to distinguish residual garbage; 3D comparison calculates height difference and center deviation based on coordinate data, with accuracy controlled at the centimeter level; Presenting quantitative results, such as garbage residual area accounting for 3%, height difference of 2.2 meters, and coordinate deviation of 0.5 meters; If it is determined to be substandard, supplementary rectification suggestions are required, such as arranging the cleanup execution unit Unit-012 to re-cleanup the residual garbage in the northwest corner of Area 1 within 24 hours, and finally forming a standardized cleanup effect verification report with complete structure and sufficient data support.

[0089] The cleanup effectiveness verification report is sent back to the dispatch node of the cleanup task instruction set to complete the management loop. Specifically, this includes: sending the standardized cleanup effectiveness verification report back to the dispatch node of the cleanup task instruction set in real time through a preset encrypted data interface (using SSL encryption technology to prevent the leakage of core information such as coordinates and residual data in the report). After the report is sent back, the system automatically verifies the integrity of the report, such as whether it contains image comparison screenshots and 3D coordinate data tables. If any are missing, a retransmission notification is triggered to ensure that the report is received completely. After receiving the report, the management system of the task dispatch node automatically associates and archives it with the full-process data corresponding to the task. The associated content includes the cleanup task instruction set (original task requirements), execution data (dispatch records, real-time tracking results, and anomaly handling records), and review data (original records of images and 3D coordinates), forming a complete task data archive. At the same time, the system automatically marks the completion of the management loop in the task status, officially ending the full-process management from task generation, through execution unit matching, task dispatch, process tracking, progress and anomaly handling, to effectiveness verification.

[0090] This embodiment objectively and accurately verifies the cleaning effect through multi-dimensional verification of data collection and comparative analysis, eliminating the possibility of superficial completion of work; the return of the cleaning effect verification report and the archiving of the entire process data complete the management loop of sanitation operations, ensuring that each task has clear results feedback; the complete effect verification data provides reliable data support for subsequent optimization of cleaning plans, adjustment of operation strategies, and improvement of the overall sanitation operation quality.

[0091] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the system as described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0092] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the system as described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0093] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A closed-loop management system for the entire process of urban sanitation waste based on drone collaborative patrols, characterized in that: include: The acquisition module is used to acquire initial image data and initial three-dimensional spatial coordinate data within the target area, and to perform image enhancement processing on the image data to obtain the image to be analyzed. The identification module is used to identify garbage targets in the image to be analyzed, and to obtain garbage targets, their visual attributes and geographical location information. The sampling module is used to perform cluster analysis on the geographical coordinates of waste targets using a density clustering algorithm, aggregating spatially adjacent waste targets into waste accumulation areas. Based on the waste accumulation areas, a convex hull boundary is constructed for each waste accumulation area. Starting from the geometric center of the convex hull boundary, five spatial sampling points are set along the main distribution direction of the convex hull boundary. The first sampling point is located at the geometric center of the convex hull boundary, the second sampling point is located at half the distance from the geometric center of the convex hull boundary to the convex hull boundary, the third sampling point is located on the convex hull boundary, the fourth sampling point is located one meter away from the convex hull boundary on the extension line of the main distribution direction outside the convex hull boundary, and the fifth sampling point is located two meters away from the convex hull boundary on the extension line of the main distribution direction outside the convex hull boundary. By calculating the spatial coordinate data distribution density of the first, second, and third sampling points, the accumulation density data of the waste accumulation area is obtained. By analyzing the variation pattern of the spatial coordinate data of the first, second, third, fourth, and fifth sampling points, the diffusion trend data of the waste accumulation area is obtained. The evaluation module is used to construct a multi-dimensional evaluation matrix based on the color distribution features, texture complexity features, and contour features in the visual attributes of the waste target, the diffusion rate and diffusion range in the diffusion trend data of the waste accumulation area, and the density per unit area in the accumulation density data. According to the multi-dimensional evaluation matrix, the waste accumulation state is divided into three levels: light accumulation state, moderate accumulation state, and heavy accumulation state according to the preset accumulation state level classification parameters, so as to obtain the accumulation state of the waste target. The calculation module is used to integrate the accumulation status of waste targets with geographical location information to determine the urgency sequence of disposal for each waste accumulation point and the collection task instruction set, where the assignment operation of the collection task instruction set corresponds to an assignment node. The dispatch module is used to assign corresponding cleaning execution units to the cleaning task instruction set, and to perform status tracking and task progress management of the cleaning process. The verification module is used to verify the cleaning effect after the cleaning task is completed, and to send the verification information back to the cleaning task instruction set dispatch node to complete the management closed loop.

2. The urban sanitation waste closed-loop management system based on UAV collaborative patrol as described in claim 1, characterized in that, Acquire initial image data and initial 3D spatial coordinate data within the target area, and perform image enhancement processing on the image data to obtain the image to be analyzed, including: Multiple drones conduct coordinated patrols of the target area according to preset flight paths, collecting initial image data and initial three-dimensional spatial coordinate data; The initial image data is subjected to illumination equalization and noise filtering to obtain a uniform quality image for analysis.

3. The urban sanitation waste closed-loop management system based on UAV collaborative patrol as described in claim 2, characterized in that, The image to be analyzed is processed for garbage target identification to obtain garbage targets, their visual attributes, and geographical location information, including: Multi-scale feature analysis is performed on the image to be analyzed with uniform quality to extract the color features, texture features and shape features of the target in the image to be analyzed; A garbage target recognition model is established based on the color, texture, and shape features of the target. By comparing the color, texture, and shape features of the target with a preset garbage feature library, scattered garbage targets are distinguished from background objects such as vegetation and buildings. The visual attribute information of the identified garbage targets and the corresponding geographical location coordinates of the garbage targets are recorded.

4. The urban sanitation waste closed-loop management system based on UAV collaborative patrol as described in claim 3, characterized in that, The accumulation status of waste targets is integrated with geographical location information to determine the urgency sequence of disposal for each waste accumulation point and the set of collection and removal task instructions. Each collection and removal task instruction set corresponds to a collection node, including: Based on the accumulation status of waste targets, a multi-factor evaluation matrix is ​​constructed, and each waste accumulation point is sorted in descending order of disposal urgency to generate a disposal urgency sequence. Based on the urgency sequence of the situation, a set of cleaning task instructions is automatically generated, which includes cleaning priority parameters, cleaning volume estimation data, and cleaning route planning data, and then sent to the task assignment node.

5. The urban sanitation waste closed-loop management system based on UAV collaborative patrol as described in claim 4, characterized in that, Assign corresponding cleanup execution units to the cleanup task instruction set, and perform status tracking and task progress management of the cleanup process, including: Based on the volume estimation and route planning in the collection task instruction set, a collection execution unit with corresponding collection capacity and scope is matched. The waste removal task instructions are assigned to the matching waste removal execution units, and a correspondence between waste removal tasks and waste removal execution units is established. By combining the real-time location data and operation data uploaded by the cleaning and transportation execution unit with the patrol data updated by the drone, the cleaning and transportation process is tracked in real time, and the real-time status tracking results are obtained. Based on real-time status tracking results, the progress of the waste removal task is dynamically updated, and warnings and handling are provided for abnormal situations.

6. The urban sanitation waste closed-loop management system based on UAV collaborative patrol as described in claim 5, characterized in that, After the waste removal task is completed, the removal effect is verified, and the verification information is sent back to the waste removal task instruction set dispatch node to complete the management loop, including: After the cleanup task progress data indicates that the cleanup task has been completed, start the data collection for the cleaned area to obtain the cleanup image data and the three-dimensional spatial coordinate data of the cleanup. The verification image data after the cleanup is compared and analyzed with the initial image data, and the verification three-dimensional spatial coordinate data after the cleanup is compared and analyzed with the initial three-dimensional spatial coordinate data to generate a cleanup effect verification report. The cleaning effect verification report is sent back to the dispatch node of the cleaning task instruction set to complete the management loop.

7. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to perform the system as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, performs the system as described in any one of claims 1 to 6.

Images