Urban garbage removal and output prediction method and system based on deep learning

Abstract

The invention discloses an urban garbage removal and output prediction method and system based on deep learning, and the method comprises the following steps: S1, obtaining a road video image, analyzing the garbage distribution and density, predicting the garbage distribution evolution, and estimating the garbage amount of a cleaning region; s2, according to the estimated garbage amount and sweeping area distribution, a sanitation vehicle is dispatched for sweeping, a video image in sweeping is collected, and the rotating speed and the vehicle speed of a sweeping disc of the vehicle are adjusted; and S3, the difference between the actual garbage amount and the estimated garbage amount is analyzed and used for correcting the estimated garbage amount. Intelligent management of environmental sanitation work is achieved, garbage distribution and cleaning amount can be efficiently estimated, a scheduling scheme and cleaning parameters can be dynamically adjusted according to actual conditions, the cleaning efficiency and quality are greatly improved, meanwhile, through data analysis and model optimization, the accuracy and reliability of the system are enhanced, and the cleaning efficiency and quality are improved. The improvement of the urban environmental sanitation level and the reasonable utilization of resources are effectively promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of garbage management, and particularly to a method and system for urban garbage removal and generation prediction based on deep learning. Background Art

[0002] With the acceleration of the urbanization process, the problem of urban garbage has become increasingly severe, having a significant impact on urban environmental sanitation and the quality of residents' lives. Traditional urban garbage management methods, such as manual inspections and regular fixed-point cleaning, are no longer able to meet the current needs of urban garbage disposal.

[0003] Currently, some technologies have attempted to solve the problem of urban garbage management by introducing intelligent means. For example, some cities have adopted a garbage collection management system based on GPS positioning and GIS technology, achieving real-time monitoring and scheduling of sanitation vehicles. However, these systems mainly rely on the physical location of the vehicles and preset cleaning routes, lacking the ability to perceive and predict the distribution and density of garbage in real time, resulting in low cleaning efficiency and serious waste of resources.

[0004] Therefore, it is necessary to improve the deficiencies in the existing technology to solve the above problems. Summary of the Invention

[0005] The present invention overcomes the deficiencies of the existing technology and provides a method and system for urban garbage removal and generation prediction based on deep learning.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a method for urban garbage removal and generation prediction based on deep learning, comprising the following steps:

[0007] S1. Obtain video images of the road surface in the cleaning area, analyze the garbage distribution and density on the road surface in the images, predict the evolution of the garbage distribution, and estimate the garbage volume of each cleaning area using a deep learning model;

[0008] S2. According to the estimated garbage volume and the cleaning area distribution, dispatch sanitation vehicles to the corresponding cleaning areas for cleaning, and collect video images during cleaning in real time to adjust the rotation speed and vehicle speed of the cleaning disks of the sanitation vehicles;

[0009] S3. Analyze the difference between the actual garbage volume and the estimated garbage volume of each cleaning area, and use the data difference to correct the estimated garbage volume.

[0010] In a preferred embodiment of the present invention, in the step of S1, the following sub-steps are included:

[0011] S11. Obtain video images of the road surface in the cleaning area and perform image processing; including noise reduction, contrast enhancement, and normalization involved;

[0012] S12. Identify the garbage in the processed image, calculate the distribution status and density of the garbage on the road surface, and generate a garbage distribution map;

[0013] S13. Combine time factors and weather conditions to make a short-term prediction of the garbage distribution and generate a predicted garbage distribution map;

[0014] S14. Based on the predicted garbage distribution map and the cleaning area map, calculate the estimated garbage volume of each cleaning area according to the garbage density and the cleaning area.

[0015] In a preferred embodiment of the present invention, in the step of S12, a deep learning model is used to generate the bounding box and class probability of the garbage, and the confidence score of the object detection

[0016] P(c|b) = Softmax(W·f(b)+b);

[0017] where f(b) is the bounding box feature; W and b are the weights and biases;

[0018] Divide the cleaning area image into N×M grids, and count the covered area of the garbage detection boxes in each grid;

[0019] Garbage distribution value of the i-th grid where A k is the detection box area of the k-th garbage in the i-th grid; A j is the total grid area;

[0020] Garbage density quantization Based on the density value ρ j , use color mapping to generate a garbage distribution heat map.

[0021] In a preferred embodiment of the present invention, in the step of S13, the time factors include seasons and time periods, and the weather conditions include rainfall, wind speed, and temperature;

[0022] Assume that the change in garbage distribution is described by a linear dynamic system:

[0023] State equation: D k = F k D k-1 + B k u k + w k ; where D k is the state vector at the k-th moment, representing the state of the garbage distribution; F k is the state transition matrix, describing the natural evolution of the state; u k is the control input vector, representing the influence of external factors on the garbage distribution; B k is the control input matrix; w kis the process noise;

[0024] Observation equation: z k = H k D k + v k ; where z k is the observation vector at the k-th moment, representing the actually observed garbage distribution; H k is the observation matrix; v k is the observation noise;

[0025] Use the Kalman filter algorithm to perform short-term prediction on the garbage distribution;

[0026] State prediction: where is the predicted state based on the estimated value at the previous moment;

[0027] Covariance prediction: where P k|k-1 is the covariance matrix of the predicted state; Q is the process noise covariance matrix;

[0028] Kalman gain calculation: where K k is the Kalman gain; R is the observation noise covariance matrix;

[0029] State update: where is the corrected state estimate;

[0030] Covariance update: P k|k = (I - K k H k )P k|k-1 where P k|k is the corrected covariance matrix; I is the identity matrix;

[0031] Based on the state vector Generate a predicted garbage distribution heat map using color mapping.

[0032] In a preferred embodiment of the present invention, in the step of S14, the garbage distribution map is superimposed on the cleaning area map, and the cleaning area is divided into N×M grids, and each grid corresponds to the garbage density value ρ j in the predicted garbage distribution map. Extract the density value of each grid from the predicted garbage distribution map and allocate the density value according to the area ratio: where A overlap is the overlapping area between the grid and the cleaning area;

[0033] For the polygon boundary of each cleaning area, calculate its actual area S k :

[0034] Among them, (x i , y i ) are the coordinates of the polygon vertices;

[0035] For each cleaning area g, accumulate the product of the density of all grids it covers and the corresponding overlapping area:

[0036] Among them, Q g is the garbage weight;

[0037] Dynamic correction factor: Considering the historical data error rate ∈, correct the estimated garbage volume:

[0038] Among them, the error rate ∈ is initially ∈ = 0 and is updated through step-by-step iteration.

[0039] In a preferred embodiment of the present invention, in the step of S2, the following sub-steps are included:

[0040] S21. Generate a sanitation vehicle scheduling plan according to the estimated garbage volume and the distribution of the cleaning areas, and dispatch the sanitation vehicles to the designated areas for cleaning;

[0041] S22. Real-time collect video images during the cleaning process, analyze the garbage density and the cleaning effect, and dynamically adjust the rotation speed and the vehicle speed of the cleaning disc of the sanitation vehicle.

[0042] In a preferred embodiment of the present invention, in the step of S21, perform dynamic division of the cleaning area priorities:

[0043] Among them, A high-density,g is the area of the high-density garbage grids in the cleaning area g; T emergency,g is the urgency of the cleaning area; α, β, and γ are weight coefficients;

[0044] Assume that the vehicle scheduling path optimization is to minimize the total cost, and the objective function is:

[0045] Among them, C ij is the driving cost of the vehicle from area i to j; x ij is a binary variable; C v is the loading capacity of vehicle v; λ is a penalty coefficient.

[0046] In a preferred embodiment of the present invention, in the step of S22, real-time locate the garbage area and output the coordinates of the detection box and the confidence level s j ;

[0047] Garbage density quantization Among them, A jis the area of the detection box; A frame is the total area of the current frame;

[0048] The cleaning effect evaluation adopts a residual network evaluation model:

[0049] Input: the current frame V t and the previous frame V t-Δt of the differential image ΔV;

[0050] Output: the cleaning effect score E t ∈[0,1], 1 indicates complete cleaning;

[0051] Loss function: combining mean squared error and structural similarity:

[0052] L = α·MSE(ΔV,Δ^V)+β·(1 - SSIM(ΔV,Δ^V)); where, α + β = 1, set by cross - validation;

[0053] Dynamic control parameter adjustment: maximizing cleaning efficiency where, P total is the sum of the cleaning disk power and vehicle energy consumption;

[0054] Constraint conditions:

[0055] Reinforcement learning strategy: generating the optimal combination of rotational speed r and vehicle speed υ:

[0056] where, the state s=(ρ t ,E t ); the action a=(υ,r); the reward R = η·E t ;

[0057] Implementation control of sanitation vehicle parameters: where, υ base and r base are the reference parameters; k υ and k r are the sensitivity coefficients.

[0058] In a preferred embodiment of the present invention, in the step of S3, the following sub - steps are included:

[0059] S31. Obtain the actual garbage amount of each cleaning area, and calculate the difference value between the actual garbage amount and the estimated garbage amount of each cleaning area;

[0060] S32. When the difference value exceeds the difference threshold, perform data correction on the estimated garbage amount;

[0061] Define the error rate ∈ as the exponentially weighted average of historical prediction errors, used to quantify the model prediction deviation:

[0062] ∈ = α g ·∈ t-1 +(1 - α g )·δ t ; where δ t is the relative difference rate at the current time step, α g is the attenuation factor used to control the weight of historical errors.

[0063] The present invention provides a system for a method of predicting urban garbage clearance and generation amount based on deep learning, including:

[0064] An image acquisition module for acquiring video images of the road surface in the cleaning area;

[0065] An analysis and evolution module for analyzing the garbage distribution and density on the road surface in the image and predicting the evolution of the garbage distribution;

[0066] A garbage amount estimation module for estimating the garbage amount in each cleaning area;

[0067] A vehicle scheduling module for outputting a scheduling and cleaning plan for sanitation vehicles according to the estimated garbage amount and the cleaning area distribution;

[0068] An image collection module for collecting video images during the cleaning of sanitation vehicles;

[0069] A dynamic adjustment module for dynamically adjusting the rotation speed and vehicle speed of the cleaning disc during the cleaning of sanitation vehicles;

[0070] A difference analysis and correction module for analyzing the difference between the actual garbage amount and the estimated garbage amount in each cleaning area and performing difference correction on the estimated garbage amount.

[0071] The present invention solves the defects existing in the background technology and has the following beneficial effects:

[0072] (1) The present invention provides a method and system for predicting urban garbage clearance and generation amount based on deep learning. Through video image analysis, intelligent scheduling, and real-time feedback optimization, it realizes the intelligent management of sanitation work. It can not only efficiently estimate the garbage distribution and cleaning amount, but also dynamically adjust the scheduling plan and cleaning parameters according to the actual situation, thereby greatly improving the cleaning efficiency and quality. At the same time, through data analysis and model optimization, it enhances the accuracy and reliability of the system, provides comprehensive and accurate technical support for urban sanitation management, and effectively promotes the improvement of urban environmental sanitation level and the rational utilization of resources.

[0073] (2) In the present invention, by identifying and classifying the garbage in the target road surface video image, analyzing its distribution, density, and dynamically predicting the garbage distribution, it can more accurately reflect the dynamic changes of the garbage, realizing real-time and accurate perception of the garbage distribution and density, providing a reliable data basis for estimation and scheduling, and significantly improving the accuracy and response speed of garbage management.

[0074] (3) In the present invention, by generating a cleaning scheduling plan for sanitation vehicles through intelligent scheduling and adjusting the cleaning parameters in real time, it can flexibly adjust the cleaning plan according to the actual situation, avoiding the blindness and resource waste of traditional fixed-route cleaning, not only improving the cleaning efficiency, but also ensuring the uniformity and consistency of the cleaning quality.

[0075] (4) In the present invention, by analyzing the difference between the actual cleaning effect and the estimated cleaning amount and data optimization, a closed-loop implementation strategy is formed, which can continuously learn and improve the prediction and scheduling strategies, helping the system to gradually improve its own accuracy and reliability, and ensuring long-term stable cleaning effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings;

[0077] Figure 1 is a flowchart of a method for predicting the generation amount and clearing of urban garbage based on deep learning according to the present invention;

[0078] Figure 2 is a structural diagram of a system for predicting the generation amount and clearing of urban garbage based on deep learning according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0079] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0080] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited by the specific embodiments disclosed below.

[0081] As shown Figure 1 in the figure, a method for predicting the amount of urban waste removal and generation based on deep learning includes the following steps:

[0082] S1. Obtain video images of the road surface in the cleaning area, analyze the waste distribution and density on the road surface in the images, predict the evolution of waste distribution, and use a deep learning model to estimate the amount of waste in each cleaning area;

[0083] S2. According to the estimated waste amount and the cleaning area distribution, dispatch sanitation vehicles to the corresponding cleaning areas for cleaning, and collect video images during cleaning in real time to adjust the rotation speed and vehicle speed of the cleaning disks of the sanitation vehicles;

[0084] S3. Analyze the difference between the actual waste amount and the estimated waste amount in each cleaning area, and use the data difference to correct the estimated waste amount.

[0085] It should be noted that through video image analysis, intelligent dispatching, and real-time feedback optimization, the present invention realizes the intelligent management of sanitation work. It can not only efficiently estimate waste distribution and cleaning volume, but also dynamically adjust the dispatching plan and cleaning parameters according to the actual situation, thereby greatly improving the cleaning efficiency and quality. At the same time, through data analysis and model optimization, the accuracy and reliability of the system are enhanced, providing comprehensive and accurate technical support for urban sanitation management, effectively promoting the improvement of urban environmental sanitation level and the rational utilization of resources.

[0086] In some specific implementation schemes, in the step of S1, it includes the following sub-steps:

[0087] S11. Obtain video images of the road surface in the cleaning area and perform image processing;

[0088] S12. Identify the waste in the processed image, calculate the distribution status and density of the waste on the road surface, and generate a waste distribution map;

[0089] S13. Combine time factors and weather conditions to perform short-term prediction on waste distribution and generate a predicted waste distribution map;

[0090] S14. Based on the predicted waste distribution map and the cleaning area map, calculate the estimated waste amount in each cleaning area according to the waste density and the cleaning area.

[0091] It should be noted that by identifying and classifying the waste in the video images of the target road surface, analyzing its distribution, density, and dynamically predicting the waste distribution, it can more accurately reflect the dynamic changes of the waste, realize the real-time and accurate perception of waste distribution and density, provide a reliable data basis for estimation and dispatching, and significantly improve the accuracy and response speed of waste management.

[0092] In this embodiment, in step S11, image processing includes noise reduction, contrast enhancement, and normalization; for noise reduction, one of Gaussian filtering, mean filtering, or median filtering is used, such as Gaussian filtering, where the kernel size (e.g., 3x3, 5x5, etc.) and standard deviation (controlling the smoothness) of the filter are set, and then the filter is applied to the image, and it is checked whether the noise is reduced and whether important information (such as edges, details, etc.) is retained to ensure the image quality; for contrast enhancement, one of histogram equalization, adaptive histogram equalization, or contrast stretching is used, such as histogram equalization, which adjusts the histogram distribution of the image to enhance the contrast, making the gray value distribution of the image more uniform and improving the contrast; for normalization, image size normalization or gray value normalization is used, such as gray value normalization, which normalizes the gray values of the image to a fixed range (e.g., [0,1] or [-1,1]) to ensure the comparability of gray values between different images.

[0093] In this embodiment, in step S12, a deep learning model is used to generate the bounding box and class probability of the garbage, and the confidence score P(c|b) of object detection is P(c|b) = Soft max(W·f(b)+b);

[0094] where f(b) is the bounding box feature; W and b are the weights and biases;

[0095] The cleaning area image is divided into a grid of N×M, and the coverage area of the garbage detection box in each grid is counted;

[0096] The garbage distribution value of the i-th grid

[0097] where A k is the detection box area of the k-th garbage in the i-th grid; A j is the total grid area;

[0098] Garbage density quantization

[0099] Based on the density value ρ j , a color mapping (such as high density in red and low density in green) is used to generate a garbage distribution heat map.

[0100] In this embodiment, in step S13, the time factors include seasons (spring, summer, autumn, winter) and time periods (morning, noon, evening), and the weather conditions include rainfall, wind speed, and temperature;

[0101] Suppose the change in garbage distribution is described by a linear dynamic system:

[0102] State equation: D k = F k D k-1 + B k uk +w k ;

[0103] where D k is the state vector at the k-th moment, representing the state of the garbage distribution; F k is the state transition matrix, describing the natural evolution of the state; u k is the control input vector, representing the influence of external factors (such as weather conditions) on the garbage distribution; B k is the control input matrix; w k is the process noise (Gaussian white noise);

[0104] Observation equation: z k = H k D k + v k ;

[0105] where z k is the observation vector at the k-th moment, representing the actually observed garbage distribution; H k is the observation matrix; v k is the observation noise (Gaussian white noise);

[0106] Use the Kalman filter algorithm to make short-term predictions for the garbage distribution;

[0107] State prediction: where is the predicted state based on the previous moment's estimated value;

[0108] Covariance prediction: where P k|k-1 is the covariance matrix of the predicted state; Q is the process noise covariance matrix;

[0109] Kalman gain calculation: where K k is the Kalman gain; R is the observation noise covariance matrix;

[0110] State update: where is the corrected state estimate;

[0111] Covariance update: P k|k = (I - K k H k ) P k|k-1 where P k|k is the corrected covariance matrix; I is the identity matrix;

[0112] Based on the state vector Generate a predicted garbage distribution heat map using color mapping (such as red for high density and green for low density).

[0113] In this embodiment, in step S14, the garbage distribution map is overlaid with the cleaning area map, and the cleaning area is divided into a grid of N×M, and each grid corresponds to the garbage density value ρ in the predicted garbage distribution map. j , and the density value of each grid is extracted from the predicted garbage distribution map, and the density value is allocated according to the area ratio: where A overlap is the overlapping area between the grid and the cleaning area;

[0114] For the polygon boundary of each cleaning area, its actual area S is calculated k :

[0115]

[0116] where (x i , y i ) are the coordinates of the polygon vertices;

[0117] For each cleaning area g, the product of the density of all the grids it covers and the corresponding overlapping area is accumulated:

[0118] where Q g is the garbage weight;

[0119] Dynamic correction factor: Considering the historical data error rate ∈, the estimated garbage volume is corrected:

[0120] where, at the initial stage, ∈ = 0, and it is updated by gradual iteration.

[0121] In some specific implementation schemes, in step S2, the following sub-steps are included:

[0122] S21. Generate a sanitation vehicle scheduling plan according to the estimated garbage volume and the distribution of the cleaning area, and dispatch the sanitation vehicle to the designated area for cleaning;

[0123] S22. Real-time collect video images during the cleaning process, analyze the garbage density and cleaning effect, and dynamically adjust the rotation speed and vehicle speed of the cleaning disc of the sanitation vehicle.

[0124] It should be noted that by generating a cleaning scheduling plan for sanitation vehicles through intelligent scheduling and adjusting the cleaning parameters in real time, the cleaning plan can be flexibly adjusted according to the actual situation, avoiding the blindness and resource waste of traditional fixed-route cleaning, not only improving the cleaning efficiency, but also ensuring the uniformity and consistency of the cleaning quality.

[0125] In this embodiment, in step S21, the dynamic division of the cleaning area priority is carried out:

[0126]

[0127] Among them, A high-density,g is the area of the high-density garbage grid within the cleaning area g; T emergency,g is the urgency level of the cleaning area (such as around transportation hubs and schools, with a value range of 0 - 1); α, β, and γ are weight coefficients (fitted through the historical task completion rate, satisfying α + β + γ = 1);

[0128] Assume that the vehicle scheduling path optimization aims to minimize the total cost, and the objective function is:

[0129]

[0130] Among them, C ij is the driving cost of the vehicle from area i to j; x ij is a binary variable (1 indicates that the vehicle travels from i to j, otherwise 0); C v is the loading capacity of vehicle v; λ is a penalty coefficient (used to handle the overloading risk).

[0131] In this embodiment, in the step of S22, the video images during the cleaning process are collected through the high-definition camera on the vehicle, and the same image processing method as in the step of S11 is adopted, which will not be elaborated here in detail;

[0132] Real-time locate the garbage area and output the detection box coordinates and confidence level s j ;

[0133] Garbage density quantization Among them, A j is the area of the detection box; A frame is the total area of the current frame;

[0134] The cleaning effect is evaluated using a residual network evaluation model:

[0135] Input: The current frame V t and the differential image ΔV of the previous frame V t-Δt ;

[0136] Output: The cleaning effect score E t ∈[0, 1], where 1 represents complete cleaning;

[0137] Loss function: Combining the mean squared error (MSE) and structural similarity (SSIM):

[0138] L = α·MSE(ΔV, Δ^V) + β·(1 - SSIM(ΔV, Δ^V)); where α + β = 1, set through cross-validation;

[0139] Dynamic control parameter adjustment: Maximize the cleaning efficiency Among them, P totalis the sum of the sweeping disk power and the vehicle energy consumption;

[0140] Constraints:

[0141] Reinforcement learning strategy: Generate the optimal combination of the rotational speed r and the vehicle speed υ:

[0142] where the state s = (ρ t , E t ); the action a = (υ, r); the reward R = η·E t ;

[0143] Implementation control of the sanitation vehicle parameters: where υ base and r base are the reference parameters; k υ and k r are the sensitivity coefficients.

[0144] In some specific implementation schemes, in the step of S3, the following sub-steps are included:

[0145] S31. Obtain the actual garbage amount of each cleaning area, and calculate the difference value between the actual garbage amount and the estimated garbage amount of each cleaning area;

[0146] S32. When the difference value exceeds the difference threshold, perform data correction on the estimated garbage amount.

[0147] It should be noted that by analyzing the difference between the actual cleaning effect and the estimated cleaning amount and data optimization, a closed-loop implementation strategy is formed, which can continuously learn and improve the prediction and scheduling strategies, helping the system to gradually improve its own accuracy and reliability, and ensuring a long-term stable cleaning effect.

[0148] In this implementation manner, in the step of S31, the actual garbage amount Q′ of each cleaning area g is obtained by weighing, and the difference value is calculated

[0149] In this implementation manner, in the step of S32, define the error rate ∈ as the exponentially weighted average of the historical prediction error, which is used to quantify the model prediction deviation:

[0150] ∈ = α g ·∈ t-1 +(1 - α g )·δ t ; where δ t is the relative difference rate at the current time step, α g is the attenuation factor, which is used to control the weight of the historical error.

[0151] As shown Figure 2 in the figure, a system for a method of predicting the amount of urban garbage removal and generation based on deep learning includes:

[0152] An image acquisition module for acquiring video images of the road surface in the cleaning area;

[0153] An analysis and evolution module for analyzing the garbage distribution and density on the road surface in the image and predicting the evolution of the garbage distribution;

[0154] A garbage volume estimation module for estimating the amount of garbage in each cleaning area;

[0155] A vehicle scheduling module for outputting a scheduling and cleaning plan for sanitation vehicles according to the estimated garbage volume and the distribution of cleaning areas;

[0156] An image collection module for collecting video images during the cleaning of sanitation vehicles;

[0157] A dynamic adjustment module for dynamically adjusting the rotation speed and vehicle speed of the cleaning disk during the cleaning of sanitation vehicles;

[0158] A difference analysis and correction module for analyzing the difference between the actual garbage volume and the estimated garbage volume in each cleaning area and correcting the difference in the estimated garbage volume.

[0159] It should be noted that the urban garbage removal and generation prediction system can implement the steps in the urban garbage removal and generation prediction method in the above-mentioned embodiments and can achieve the same technical effects. Refer to the description in the above-mentioned embodiments and details will not be elaborated here.

[0160] Based on the ideal embodiments of the present invention as an inspiration, through the above description, for those skilled in the art, it is obvious that the present invention is not limited to the details of the above-mentioned exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. Therefore, in any aspect, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

[0161] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for predicting the amount of urban waste removal and generation based on deep learning, characterized in that, It includes the following steps: S1. Obtain the video images of the road surface in the cleaning area, analyze the garbage distribution and density on the road surface in the images, predict the evolution of the garbage distribution, and use a deep learning model to estimate the garbage volume in each cleaning area; S2. According to the estimated garbage volume and the cleaning area distribution, dispatch sanitation vehicles to the corresponding cleaning areas for cleaning, and collect the video images during cleaning in real time to adjust the rotation speed and vehicle speed of the cleaning disks of the sanitation vehicles; S3. Analyze the difference between the actual garbage volume and the estimated garbage volume in each cleaning area, and use the data difference to correct the estimated garbage volume.

2. The method for predicting the urban waste removal and generation amount based on deep learning according to claim 1, wherein: In the step of S1, it includes the following sub-steps: S11. Obtain the video images of the road surface in the cleaning area and perform image processing; including noise reduction, contrast enhancement, and normalization involved; S12. Identify the garbage in the processed image, calculate the distribution status and density of the garbage on the road surface, and generate a garbage distribution map; S13. Combine the time factor and weather conditions to make a short-term prediction of the garbage distribution and generate a predicted garbage distribution map; S14. Based on the predicted garbage distribution map and the cleaning area map, calculate the estimated garbage volume in each cleaning area according to the garbage density and the cleaning area.

3. The method for predicting the urban garbage removal and generation amount based on deep learning according to claim 2, wherein: In the step of S12, use a deep learning model to generate the bounding box and class probability of the garbage, and the confidence score P(c|b) of object detection = Soft max(W·f(b)+b); where f(b) is the bounding box feature; W and b are weights and biases; Divide the cleaning area image into a grid of N×M, and count the covered area of the garbage detection boxes in each grid; Garbage distribution value of the i-th grid where A k is the detection box area of the k-th piece of garbage in the i-th grid; A j is the total area of the grid; Garbage density quantification Based on the density value ρ j A heat map of garbage distribution is generated using color mapping.

4. A method for predicting the amount of urban waste removal and generation based on deep learning according to claim 2, characterized in that: In the step of S13, the time factor includes season and time period, and the weather conditions include rainfall, wind speed, and temperature; Assume that the change of the garbage distribution is described by a linear dynamic system: State equation: D k = F k D k-1 + B k u k + w k ; Among them, D k is the state vector at the k-th moment, representing the state of the garbage distribution; F k is the state transition matrix, describing the natural evolution of the state; u k is the control input vector, representing the influence of external factors on the garbage distribution; B k is the control input matrix; w k is the process noise; Observation equation: z k = H k D k + v k ; where z k is the observation vector at the k-th moment, representing the actually observed garbage distribution; H k is the observation matrix; v k is the observation noise; Use the Kalman filter algorithm to make a short-term prediction of the garbage distribution; State prediction: wherein, is the predicted state based on the estimated value at the previous moment; Covariance prediction: where P k|k-1 is the covariance matrix of the predicted state; Q is the process noise covariance matrix; Kalman gain calculation: where K k is the Kalman gain; R is the observation noise covariance matrix; Status update: Among them, is the corrected state estimate; Covariance update: P kk = (I - K k H k )P kk-1 where P kk is the corrected covariance matrix; I is the identity matrix; Based on the state vector Use color mapping to generate a heat map of the predicted garbage distribution.

5. The method for predicting the urban waste removal and generation amount based on deep learning according to claim 2, characterized in that: In the step of S14, the garbage distribution map is overlaid with the cleaning area map, and the cleaning area is divided into a grid of N×M, and each grid corresponds to the garbage density value ρ in the predicted garbage distribution map j , extract the density value of each grid from the predicted garbage distribution map, and allocate the density value according to the area ratio: where A overlap is the overlapping area between the grid and the cleaning area; For the polygon boundary of each cleaning area, calculate its actual area S k : Among them, (x i , y i ) are the coordinates of the polygon vertices; For each cleaning area g, accumulate the product of the density of all grids it covers and the corresponding overlapping area: Among them, Q g is the garbage weight; Dynamic correction factor: Consider the historical data error rate ∈ and correct the garbage estimation amount: Among them, the error rate ∈ is initially ∈ = 0 and is updated through step-by-step iteration.

6. The method for predicting the urban garbage removal and generation amount based on deep learning according to claim 1, wherein: In the step of S2, it includes the following sub-steps: S21. According to the estimated garbage volume and the cleaning area distribution, generate a sanitation vehicle dispatch plan and dispatch sanitation vehicles to the designated areas for cleaning; S22. Collect the video images during the cleaning process in real time, analyze the garbage density and cleaning effect, and dynamically adjust the rotation speed and vehicle speed of the cleaning disks of the sanitation vehicles.

7. A method for predicting the amount of urban waste removal and generation based on deep learning according to claim 6, characterized in that: In the step of S21, perform dynamic division of the cleaning area priority: Among them, A high-density,g is the area of the high-density garbage grid within the cleaning area g; T emergency,g is the urgency of the cleaning area; α, β, and γ are weighting coefficients; Assume that the vehicle dispatch path optimization is to minimize the total cost, and the objective function is: Among them, C ij is the driving cost of the vehicle from area i to j; x ij is a binary variable; C v is the loading capacity of vehicle v; λ is the penalty coefficient.

8. A method for predicting the amount of urban waste removal and generation based on deep learning according to claim 6, characterized in that: In the step of S22, the garbage area is located in real time, and the coordinates of the detection frame and the confidence level s are output j ; Garbage density quantification where A j is the area of the detection box; A frame is the total area of the current frame; The cleaning effect evaluation uses a residual network evaluation model: Input: Current frame V t And the previous frame V t-Δt The differential image ΔV; Output: Cleaning effect score E t ∈[0, 1], where 1 represents complete cleaning; Loss function: Combine the mean square error and structural similarity: L = α·MSE(ΔV,Δ^V)+β·(1 - SSIM(ΔV,Δ^V)); where α + β = 1, set by cross-validation; Dynamic control parameter adjustment: maximizing cleaning efficiency Among them, P total is the sum of the cleaning disk power and the vehicle energy consumption; Constraints: Reinforcement learning strategy: Generate the optimal combination of the rotation speed r and the vehicle speed υ: Among them, the state s = (ρ t , E t ); the action a = (υ, r); the reward R = η·E t ; Implementation control of sanitation vehicle parameters: Among them, υ base and r base are reference parameters; k υ and k r are sensitivity coefficients.

9. A method for predicting the amount of urban waste removal and generation based on deep learning according to claim 1, characterized in that: In the step of S3, it includes the following sub-steps: S31. Obtain the actual garbage volume in each cleaning area and calculate the difference value between the actual garbage volume and the estimated garbage volume in each cleaning area; S32. When the difference value exceeds the difference threshold, perform data correction on the estimated garbage volume; Define the error rate ∈ as the exponentially weighted average of historical estimation errors, which is used to quantify the prediction deviation of the model: ∈ = α g ·∈ t-1 +(1 - α g )·δ t ; where δ t is the relative difference rate at the current time step, α g is the decay factor used to control the weight of historical errors.

10. A system for a method of predicting the amount of urban waste removal and generation based on deep learning according to any one of claims 1-9, characterized in that, Including: An image acquisition module, which is used to acquire video images of the road surface in the cleaning area; An analysis and evolution module, which is used to analyze the garbage distribution and density on the road surface in the image and predict the evolution of the garbage distribution; A garbage volume estimation module, which is used to estimate the garbage volume of each cleaning area; A vehicle scheduling module, which is used to output the scheduling and cleaning plan of the sanitation vehicle according to the estimated garbage volume and the cleaning area distribution; An image collection module, which is used to collect video images during the cleaning of the sanitation vehicle; A dynamic adjustment module, which is used to dynamically adjust the rotation speed and vehicle speed of the cleaning disk during the cleaning process of the sanitation vehicle; A difference analysis and correction module, which is used to analyze the difference between the actual garbage volume and the estimated garbage volume of each cleaning area and perform difference correction on the estimated garbage volume.

Images