Garbage sorting equipment and sorting system based on visual identification

Abstract

The invention provides garbage sorting equipment and sorting system based on visual identification, and relates to the technical field of intelligent garbage sorting, the garbage sorting equipment comprises the sorting system and the sorting equipment, the sorting system comprises a visual identification module, a control module, a mechanical claw driving module, a garbage compression module, a cleaning module, a collection barrel control module and a parameter optimization module, in the aspect of sorting accuracy, the garbage type and position can be accurately recognized through a combined algorithm of a convolutional neural network, a background difference method, a support vector machine and a local binary pattern in the visual recognition module, and the sorting precision is greatly improved. The control module adopts the combination of a fuzzy control algorithm and a self-adaptive PID (Proportion Integration Differentiation) algorithm, so that the coordinated operation of each module is more accurate, the smooth and efficient whole sorting process is ensured, and the mechanical claw driving module effectively plans the path of a mechanical claw and ensures proper grabbing force by applying an A * algorithm, a genetic algorithm, a model prediction control algorithm and a force feedback control algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent garbage sorting, in particular to a garbage sorting device and system based on visual recognition. BACKGROUND

[0002] In today's society, the process of urbanization is accelerating, the size of the city is expanding, and the population is gathering. This development trend, while driving economic prosperity and social progress, also inevitably brings a series of environmental problems, among which the sharp increase in garbage production is particularly prominent. With the improvement of people's living standards, the consumption pattern is becoming increasingly diversified, and the types and components of garbage have become increasingly complex, covering various plastic products, paper products, metal products, kitchen waste and hazardous waste, etc.

[0003] Garbage sorting, as a key link in the garbage disposal process, is self-evident. However, for a long time, the traditional garbage sorting method mainly relies on manual operation. In this mode, the sorting workers need to work in harsh environments for a long time, and face piles of garbage, which need to be classified one by one. Due to the large amount of garbage, the efficiency of manual sorting is extremely low, far from meeting the growing demand for garbage disposal. Moreover, this high-intensity labor is a great test for the physical strength and endurance of workers, and long-term engagement in this work is likely to cause workers' physical fatigue and injury, and the high labor intensity becomes an important factor restricting its sustainable development.

[0004] In recent years, automated garbage sorting technology has gradually developed, and some existing sorting equipment uses simple mechanical structures and sensors to realize the classification and collection of garbage, but there are many problems: (1) In terms of visual recognition, some systems only use a single image recognition algorithm, which has low recognition accuracy, especially for garbage feature extraction and classification in complex environments; (2) On the control module, the traditional control algorithm cannot well adapt to the complex and variable working conditions in the sorting process, resulting in poor coordination between components, affecting the sorting efficiency; (3) The driving of the mechanical claw often causes collision or unstable grabbing due to poor path planning, the garbage compression module cannot reasonably adjust the compression parameters according to the actual situation of the garbage, which is easy to cause equipment wear or poor compression effect, the cleaning module is not thorough enough for residual garbage, the collection barrel control is not precise enough, and the whole system lacks an effective parameter optimization mechanism, making it difficult to maintain high-performance sorting operations for a long time.

[0005] Therefore, a garbage sorting device and system based on visual recognition is needed to solve the above problems. SUMMARY

[0006] TECHNICAL PROBLEM SOLVED

[0007] In view of the deficiencies of the prior art, the garbage sorting equipment and sorting system based on visual recognition are provided to solve the following problems:

[0008] 1. In the aspect of visual recognition, some systems only use a single image recognition algorithm, which has low recognition accuracy, especially for garbage feature extraction and classification in complex environments.

[0009] 2. On the control module, the traditional control algorithm cannot well adapt to the complex and variable working conditions in the sorting process, resulting in insufficient precision in coordination between components and affecting sorting efficiency.

[0010] 3. The driving of the mechanical claw often collides or is unstable due to poor path planning, the garbage compression module cannot reasonably adjust the compression parameters according to the actual situation of the garbage, which easily causes equipment wear or poor compression effect, the cleaning module does not clean the residual garbage thoroughly, the collection bucket control is not precise enough, and the entire system lacks an effective parameter optimization mechanism, making it difficult to maintain high-performance sorting operations for a long time.

[0011] Technical scheme

[0012] To achieve the above purpose, the following technical solutions are used: the garbage sorting equipment and sorting system based on visual recognition, including a sorting system and a sorting device, the sorting system includes a visual recognition module, a control module, a mechanical claw driving module, a garbage compression module, a cleaning module, a collection bucket control module and a parameter optimization module;

[0013] The sorting device includes a sorting box, a control mechanism is arranged on the top of the sorting box, an installation rod is arranged in the sorting box, a holder is arranged on the surface of the installation rod, a first transmission assembly is arranged in the sorting box, a first push plate is arranged on the surface of the first transmission assembly, a second transmission assembly is arranged at the bottom of the sorting box close to the first transmission assembly, a second push plate is arranged on the surface of the second transmission assembly, a cleaning assembly is arranged on the surface of the holder, first and second collection buckets are arranged on the both sides of the bottom of the sorting box, the cleaning assembly includes a connecting plate, a third servo motor, a driving screw rod, a brush and a guide rod, the third servo motor is arranged on the surface of the connecting plate, the driving screw rod is arranged on the output end of the third servo motor, the brush is mounted on the surface of the driving screw rod, the guide rods are connected to the surfaces of the two connecting plates, one end of the brush is movably connected with the guide rods, a guide rail is arranged in the sorting box, a mechanical claw is mounted on the surface of the guide rail, first and second side plates are arranged on the both sides of the sorting box.

[0014] Preferably, the visual recognition module collects images and identifies features of the garbage falling on the surface of the holder, and provides garbage type and position information for the control module to achieve accurate sorting. The image collection adopts an algorithm combining convolutional neural network (CNN) and background difference method. The CNN is used to preliminarily extract garbage image features, and the background difference method is used to enhance the contrast between the target and the background, thereby improving the accuracy of garbage recognition. The feature recognition uses a fusion algorithm of support vector machine (SVM) and local binary pattern (LBP). The SVM classifies the extracted features, and the LBP further refines the texture features, so as to accurately determine the garbage type.

[0015] Preferably, the control module receives information from the visual recognition module, and cooperates with the mechanical claw driving module, the garbage compression module, the cleaning module and the collection barrel control module to coordinate the operation of each module and ensure the orderly work of the entire sorting system. The control module combines fuzzy control algorithm and adaptive proportional-integral-derivative (PID) algorithm. The fuzzy control algorithm adjusts the control strategy according to the visual recognition result and the state of each module, and the adaptive PID algorithm dynamically adjusts the PID parameters according to the real-time operation error of the system to accurately control the coordination of the actions of each module.

[0016] Preferably, the mechanical claw driving module controls the movement of the mechanical claw on the xyz three-axis guide rail according to the instructions of the control module, accurately grasps the garbage that needs to be sorted separately, and transfers it to the designated position. The mechanical claw driving module combines A algorithm in path planning algorithm and genetic algorithm. The A algorithm plans the preliminary path of the mechanical claw from the current position to the target garbage position, and the genetic algorithm optimizes the path to avoid collision and improve efficiency. Meanwhile, the model predictive control (MPC) and force feedback control algorithm are adopted. The MPC predicts the next action according to the mechanical claw motion model, and the force feedback control algorithm ensures that the grasping force of the mechanical claw is appropriate, so as not to damage the garbage or cause the grasping to fail.

[0017] Preferably, the garbage compression module drives the first push plate and the second push plate to move towards each other after the separate sorting is completed, so as to compress the remaining garbage, reduce the volume of the garbage and improve the storage efficiency. The garbage compression module combines finite element analysis algorithm and fuzzy logic algorithm. The finite element analysis algorithm simulates the stress deformation of the garbage in the compression process to determine the optimal compression path and force range. The fuzzy logic algorithm adjusts the compression parameters according to the type and quantity of the garbage, so as to ensure the compression effect while avoiding excessive pressure on the garbage treatment equipment. Meanwhile, the pressure control algorithm based on particle swarm optimization (PSO) and simulated annealing algorithm is used. The PSO algorithm preliminarily optimizes the compression pressure, and the simulated annealing algorithm further adjusts to prevent local optimal solution, so as to ensure the stability and efficiency of the compression process.

[0018] Preferably, the cleaning module drives the brush to move on the surface of the holder after the garbage sorting operation is completed, to clean the residual plastic pieces and tiny garbage, keep the holder clean, and ensure that the subsequent sorting operation is not affected, the cleaning module adopts a combination of a watershed algorithm and a region growing algorithm in an image segmentation algorithm, the watershed algorithm performs rough segmentation on the image of the surface of the holder, and the region growing algorithm further accurately segments the residual garbage area to determine the cleaning target; the motion control of the driven brush adopts a combination of an ant colony algorithm and a Dijkstra algorithm, the ant colony algorithm plans the approximate cleaning path of the brush, and the Dijkstra algorithm optimizes the path, so that the brush covers the entire surface of the holder in the shortest path, and efficient cleaning is realized.

[0019] Preferably, the collecting barrel control module controls the first side plate and the second side plate to rotate accurately through a motor, so as to control the exposure and storage of the first collecting barrel and the second collecting barrel, realize accurate storage of the sorted garbage into the corresponding collecting barrel, a partition plate is arranged in the first collecting barrel for classified collection of different types of garbage, the collecting barrel control module uses a neural network prediction algorithm and a Kalman filtering algorithm, the neural network prediction algorithm predicts the use of the collecting barrel according to the garbage sorting progress and type, the Kalman filtering algorithm filters and corrects the prediction result, and the prediction accuracy is improved; the motor control adopts a combination of a sliding film variable structure control algorithm and a robust control algorithm, the sliding film variable structure control algorithm ensures the rapidity and accuracy of the rotation of the side plate, and the robust control algorithm enhances the anti-interference ability of the system to the parameter changes of the motor and external interference, and ensures the accurate control of the collecting barrel.

[0020] Preferably, the parameter optimization module optimizes and adjusts the identification parameters of the visual identification module, the motion parameters of the mechanical claw driving module, the compression strength parameters of the garbage compression module, the cleaning path parameters of the cleaning module, and the rotation angle parameters of the collecting barrel control module according to the feedback information of multiple sorting operations, to improve the overall performance and sorting accuracy of the system, the parameter optimization module adopts a combination of a multi-objective genetic algorithm MOGA and a biological evolution algorithm BEA, the MOGA optimizes multiple parameters at the same time, considers the balance between multiple targets, and the BEA simulates the biological evolution process to further search for a more optimal parameter combination; and a local optimization algorithm based on a gradient descent algorithm and a trust region algorithm is combined, the gradient descent algorithm quickly finds the optimal solution direction in a local area, and the trust region algorithm guarantees the stability and convergence of the optimization process, and avoids falling into a local optimal solution.

[0021] Beneficial effects

[0022] The application provides a garbage sorting device and a sorting system based on visual identification.

[0023] 1、The present application is in the sorting accuracy, through the combination algorithm of convolution neural network in visual recognition module, background difference method, support vector machine and local binary pattern, can accurately identify the type and position of garbage, greatly improve the sorting precision. The fuzzy control algorithm and adaptive PID algorithm are combined in the control module, so that each module operates more accurately, and the whole sorting process is smooth and efficient. The A* algorithm, genetic algorithm, model predictive control and force feedback control algorithm are used in the mechanical claw driving module, which effectively plans the mechanical claw path and ensures the appropriate grabbing force, avoids collision and grabbing failure.

[0024] 2、The garbage compression module in the application utilizes finite element analysis algorithm, fuzzy logic algorithm, particle swarm optimization algorithm and simulated annealing algorithm to reasonably compress garbage, reduce volume and protect equipment. The watershed algorithm, region growing algorithm, ant colony algorithm and Dijkstra algorithm of the cleaning module are combined to completely clean the gimbal residual garbage. The collection barrel control module accurately controls the collection barrel operation through neural network prediction algorithm, Kalman filter algorithm, synovial membrane variable structure control algorithm and robust control algorithm. The multi-objective genetic algorithm, simulated biological evolution algorithm, gradient descent algorithm and trust region algorithm of the parameter optimization module are combined to continuously optimize system parameters, ensure long-term high-performance sorting, improve the quality and efficiency of garbage sorting, and reduce labor cost and equipment maintenance cost. BRIEF DESCRIPTION OF DRAWINGS

[0025] Fig. 1 The specific flowchart of the present application is shown in the figure.

[0026] Fig. 2 The internal structure diagram of the sorting box of the present application is shown in the figure.

[0027] Fig. 3 The side view of the sorting box of the present application is shown in the figure.

[0028] Fig. 4 The rear view of the sorting box of the present application is shown in the figure.

[0029] Fig. 5 The internal structure diagram of the gimbal of the present application is shown in the figure.

[0030] Fig. 6 The surface structure diagram of the gimbal of the present application is shown in the figure.

[0031] LEGEND:

[0032] 1. Sorting box; 2. Control mechanism; 3. Mounting rod; 4. Gimbal; 5. First transmission assembly; 6. First push plate; 7. Second transmission assembly; 8. Second push plate; 9. First collection bin; 10. Second collection bin; 11. Cleaning assembly; 1101. Connecting plate; 1102. Servo motor; 1103. Drive screw; 1104. Brush; 1105. Guide rod; 12. Guide rail; 13. Mechanical gripper; 14. First side plate; 15. Second side plate. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Specific Implementation Example 1:

[0035] like Figs. 1-6 As shown, the waste sorting equipment and system based on vision recognition includes a sorting system and sorting equipment. The sorting system includes a vision recognition module, a control module, a mechanical claw drive module, a waste compression module, a cleaning module, a collection bin control module, and a parameter optimization module.

[0036] The sorting equipment includes a sorting box 1, a control mechanism 2 at the top of the sorting box 1, a mounting rod 3 inside the sorting box 1, a gimbal 4 on the surface of the mounting rod 3, a first transmission assembly 5 inside the sorting box 1, a first push plate 6 on the surface of the first transmission assembly 5, a second transmission assembly 7 near the bottom of the first transmission assembly 5 inside the sorting box 1, a second push plate 8 on the surface of the second transmission assembly 7, a cleaning assembly 11 on the surface of the gimbal 4, a first collection bin 9 and a second collection bin 10 on both sides of the bottom of the sorting box 1, and the cleaning assembly 11 includes a connecting plate 1101 and a third servo motor. The sorting box 1 includes a motor 1102, a drive screw 1103, a brush 1104, and a guide rod 1105. A third servo motor 1102 is provided on the surface of the connecting plate 1101. A drive screw 1103 is provided at the output end of the third servo motor 1102. A brush 1104 is mounted on the surface of the drive screw 1103. Guide rods 1105 are connected to the surfaces of the two connecting plates 1101. One end of the brush 1104 is movably connected to the guide rod 1105. A guide rail 12 is provided inside the sorting box 1. A mechanical claw 13 is mounted on the surface of the guide rail 12. A first side plate 14 and a second side plate 15 are provided on both sides inside the sorting box 1.

[0037] The visual recognition module collects images and identifies features of the garbage falling on the surface of the holder 4, and provides the garbage type and position information for the control module to realize accurate sorting. The image collection adopts an algorithm combining convolutional neural network (CNN) and background difference method. The CNN is used to preliminarily extract the features of the garbage image, and the background difference method is used to enhance the contrast between the target and the background, thereby improving the accuracy of garbage identification. The feature recognition uses a fusion algorithm of support vector machine (SVM) and local binary pattern (LBP). The SVM classifies the extracted features, and the LBP further refines the texture features, so as to accurately determine the type of garbage.

[0038] The control module receives the information of the visual recognition module, and cooperates with the mechanical claw driving module, the garbage compression module, the cleaning module and the collection barrel control module to coordinate the operation of each module and ensure the orderly work of the whole sorting system. The control module adopts a fuzzy control algorithm combined with an adaptive proportional-integral-derivative (PID) algorithm. The fuzzy control algorithm adjusts the control strategy according to the visual recognition result and the state of each module, and the adaptive PID algorithm dynamically adjusts the PID parameters according to the real-time operation error of the system to accurately control the coordination of the actions of each module.

[0039] The mechanical claw driving module controls the movement of the mechanical claw on the xyz three-axis guide rail 12 according to the instructions of the control module, accurately grasps the garbage that needs to be sorted separately, and transfers it to the designated position. The mechanical claw driving module uses the combination of A algorithm and genetic algorithm in path planning algorithm. The A algorithm plans the preliminary path of the mechanical claw from the current position to the target garbage position, and the genetic algorithm optimizes the path to avoid collision and improve efficiency. Meanwhile, the model predictive control (MPC) and force feedback control algorithm are adopted. The MPC predicts the next action according to the mechanical claw motion model, and the force feedback control algorithm ensures that the grasping force of the mechanical claw is appropriate, so as not to damage the garbage or cause the grasping to fail.

[0040] After the separate sorting is completed, the garbage compression module drives the first push plate 6 and the second push plate 8 to move towards each other to compress the remaining garbage, reduce the volume of the garbage and improve the storage efficiency. The garbage compression module adopts the combination of finite element analysis algorithm and fuzzy logic algorithm. The finite element analysis algorithm simulates the stress deformation of the garbage in the compression process to determine the optimal compression path and force range. The fuzzy logic algorithm adjusts the compression parameters according to the type and quantity of the garbage, so as to ensure the compression effect while avoiding excessive pressure on the garbage treatment equipment. Meanwhile, the pressure control algorithm based on particle swarm optimization (PSO) and simulated annealing algorithm is used. The PSO algorithm preliminarily optimizes the compression pressure, and the simulated annealing algorithm further adjusts to prevent local optimal solution, ensuring stable and efficient compression process.

[0041] The cleaning module drives the brush 1104 to move on the surface of the gimbal 4 after the garbage sorting operation is completed, to remove the residual plastic pieces and tiny garbage, keep the gimbal 4 clean, and ensure that the subsequent sorting operation is not affected. The cleaning module combines the watershed algorithm and the region growing algorithm in the image segmentation algorithm. The watershed algorithm coarsely segments the surface image of the gimbal, and the region growing algorithm further accurately segments the residual garbage area to determine the cleaning target. The motion control of the driving brush combines the ant colony algorithm and the Dijkstra algorithm. The ant colony algorithm plans the general cleaning path of the brush, and the Dijkstra algorithm optimizes the path, so that the brush covers the entire gimbal surface with the shortest path, achieving efficient cleaning.

[0042] The collection barrel control module controls the accurate rotation of the first side plate 14 and the second side plate 15 through the motor, thereby controlling the exposure and storage of the first collection barrel 9 and the second collection barrel 10, achieving accurate storage of the sorted garbage into the corresponding collection barrel. A partition plate is arranged in the first collection barrel 9 for classified collection of different types of garbage. The collection barrel control module uses a neural network prediction algorithm and a Kalman filter algorithm. The neural network prediction algorithm predicts the usage of the collection barrel according to the garbage sorting progress and type, and the Kalman filter algorithm filters and corrects the prediction result to improve the prediction accuracy. The motor control combines a sliding film variable structure control algorithm and a robust control algorithm. The sliding film variable structure control algorithm ensures the rapidity and accuracy of the rotation of the side plate, and the robust control algorithm enhances the anti-interference ability of the system to the changes in motor parameters and external interference, ensuring the accurate control of the collection barrel.

[0043] The parameter optimization module optimizes and adjusts the identification parameters of the visual recognition module, the motion parameters of the mechanical claw driving module, the compression force parameters of the garbage compression module, the cleaning path parameters of the cleaning module, and the rotation angle parameters of the collection barrel control module, etc. according to the feedback information of multiple sorting operations, to improve the overall performance and sorting accuracy of the system. The parameter optimization module combines a multi-objective genetic algorithm MOGA and a biological evolution algorithm BEA. The MOGA optimizes multiple parameters simultaneously and considers the balance between multiple objectives. The BEA simulates the biological evolution process to further search for better parameter combinations. Meanwhile, the local optimization algorithm based on the gradient descent algorithm and the trust region algorithm is combined. The gradient descent algorithm quickly finds the optimal solution direction in the local area, and the trust region algorithm ensures the stability and convergence of the optimization process to avoid falling into a local optimal solution. Specific embodiment two:

[0045] As Figs. 1-6 shown, the following details the key algorithms mentioned in embodiment one, including their core mathematical formulas and explanations:

[0046] 1. Visual recognition module - convolutional neural network (CNN) combined with support vector machine (SVM) related formula convolutional neural network (CNN) convolutional layer

[0047] Let input image feature map I(x, y) (x and y represent coordinate positions in the image) with size m x n, convolution kernel K(x ′ ,y ′ ) with size k x k, step s, and bias b. Then the feature map O(i, j) after convolution is calculated by the formula

[0048]

[0049] This formula is the basis for CNN to extract image features. The convolution kernel slides over the input image, multiplies the pixel values at the corresponding positions, sums them up, and adds the bias to obtain each value of the output feature map. Different convolution kernels can extract different features, such as horizontal edges, vertical edges, etc.

[0050] Support Vector Machine (SVM) decision function (linear case)

[0051]

[0052] For an input feature vector x = (x1, x2, …, x d ) (where d is the feature dimension, composed of image features extracted by CNN), the decision function of SVM is: f(x) = w · x + b, where w = (w1, w2, …, w d ) is the weight vector and b is the bias term. The classification decision rule is:

[0053] SVM aims to find a hyperplane f(x) = 0 to separate different classes of data. The weight vector w determines the direction of the hyperplane, and the bias b determines the position of the hyperplane. By adjusting w and b through training data, the maximum interval between the two classes of data is obtained.

[0054] 2. Control module - fuzzy control algorithm combined with adaptive PID algorithm related formulas

[0055] Fuzzy control algorithm

[0056] Let the domain of error e be [-e m ax,e max ], divided into n fuzzy subsets A i , with membership functions adopting a triangular form

[0057]

[0058] Similarly, let the domain of error change rate Δe and fuzzy subsets be equal. The fuzzy control rule table R is a table of "if E is A i andΔE is B jthen U is C ij a matrix of rules.

[0059] The fuzzy set U of the fuzzy inference output is defuzzified by the center of gravity method to obtain the precise control output u:

[0060]

[0061] where u k is a discrete value in the output universe of discourse.

[0062] Adaptive PID algorithm

[0063] The formula of the traditional PID algorithm is:

[0064]

[0065] In the adaptive PID, the proportional coefficient K p = K p0 + f1(e, Delta e), the integral coefficient K i = K i0 + f2(e, Delta e), and the differential coefficient K d = K d0 + f3(e, Delta e). For example, f1(e, Delta e) can be f1(e, Delta e) = k1e + k2Delta e (k1 and k2 are adaptive adjustment parameters).

[0066] Fuzzy control can handle the uncertainty of complex systems through fuzzification of input, inference according to fuzzy rules, and defuzzification of output. Adaptive PID dynamically adjusts parameters according to system error and error change rate, making control more accurate, and the combination of the two can better coordinate the operation of each module. Specific embodiment three:

[0068] As Figs. 1-6 shown, in the device, the guide rail 12 is an xyz three-axis guide rail 12, which provides a moving path for the mechanical claw 13 in three-dimensional space, so that it can accurately reach any position in the sorting box 1. The holder 5 is installed on the mounting rod 4 and is used to store garbage, facilitating the operation of the subsequent mechanical claw 18. The surface of the holder 5 is driven by the servo motor 1102 to drive the driving screw 1103 to rotate, thereby making the brush 1104 move along the guide rod 1105 on the holder 4 to clean the residual plastic pieces and small garbage. The highest end of the cleaning assembly 11 is located at a horizontal plane lower than the lowest end of the first push plate 6 and the second push plate 8, so as to avoid collision between the cleaning assembly 11, the first push plate 6 and the second push plate 8. The first collection barrel 12 is provided with a partition plate to collect multiple types of garbage, and the second collection barrel 13 collects the compressed garbage.

[0069] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and implementations, it is to be understood that the terminology used is for the purpose of descriptive clarity and that it is intended to be limited only by the words recited in the appended claims. It is to be understood that the terms "including", "comprising", "consisting" and variations thereof do not preclude the addition of further integers to the combinations of integers specified in the claims. It is to be understood that the terms "including", "comprising", "consisting" and variations thereof encompass the various features of the embodiments described herein. It is not intended that the application be limited to the implementation that is described in detail and / or shown in the drawings. It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and implementations, it is to be understood that the terminology used is for the purpose of descriptive clarity and that it is intended to be limited only by the words recited in the appended claims.

[0070] While the embodiments of the application have been shown and described herein, it is to be understood that the application is not limited to the embodiments described, and it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and implementations, it is to be understood that the terminology used is for the purpose of descriptive clarity and that it is intended to be limited only by the words recited in the appended claims.

Claims

1. A waste sorting device and sorting system based on visual recognition, characterized in that: The application relates to a sorting system and a sorting device, wherein the sorting system comprises a visual recognition module, a control module, a mechanical claw driving module, a garbage compression module, a cleaning module, a collecting barrel control module and a parameter optimization module. The sorting device comprises a sorting box (1), a control mechanism (2) arranged at the top of the sorting box (1), a mounting rod (3) arranged in the sorting box (1), a holder (4) arranged on the surface of the mounting rod (3), a first transmission assembly (5) arranged in the sorting box (1), a first push plate (6) arranged on the surface of the first transmission assembly (5), a second transmission assembly (7) arranged at the bottom of the sorting box (1) and close to the first transmission assembly (5), a second push plate (8) arranged on the surface of the second transmission assembly (7), a cleaning assembly (11) arranged on the surface of the holder (4), a first collecting barrel (9) and a second collecting barrel (10) arranged at the two sides of the bottom of the sorting box (1), wherein the cleaning assembly (11) comprises a connecting plate (1101), a third servo motor (1102), a driving screw rod (1103), a brush (1104) and a guide rod (1105), the third servo motor (1102) is arranged on the surface of the connecting plate (1101), the driving screw rod (1103) is arranged on the output end of the third servo motor (1102), the brush (1104) is arranged on the surface of the driving screw rod (1103), the guide rod (1105) is arranged on the surfaces of the two connecting plates (1101), one end of the brush (1104) is movably connected with the guide rod (1105), a guide rail (12) is arranged in the sorting box (1), a mechanical claw (13) is arranged on the surface of the guide rail (12), a first side plate (14) and a second side plate (15) are arranged at the two sides of the sorting box (1), and the guide rail (12) is an xyz three-axis guide rail. 2.The visual recognition-based garbage sorting device and system according to claim 1, characterized in that: The visual recognition module collects images and identifies features of the garbage falling on the surface of the holder (4), provides garbage type and position information for the control module, and realizes accurate sorting, wherein the image collection adopts an algorithm combining a convolutional neural network (CNN) and a background difference method, the garbage image features are preliminarily extracted through the CNN, the background difference method is used to enhance the contrast between the target and the background, and the accuracy of garbage identification is improved; the feature identification adopts a support vector machine (SVM) and a local binary pattern (LBP) fusion algorithm, the SVM classifies the extracted features, and the LBP further refines the texture features, so that the garbage type can be accurately judged. 3.The visual recognition-based garbage sorting device and system according to claim 1, characterized in that: The control module receives information of the visual recognition module, and cooperates with the mechanical claw driving module, the garbage compression module, the cleaning module and the collection barrel control module to coordinate the operation of each module and ensure the orderly work of the whole sorting system. 4.The visual recognition-based garbage sorting device and system according to claim 1, characterized in that: The mechanical claw driving module controls the movement of the mechanical claw on the xyz three-axis guide rail (12) according to the instruction of the control module, accurately grasps the garbage that needs to be sorted separately, and transfers it to the designated position. The mechanical claw driving module uses the combination of A algorithm and genetic algorithm in path planning algorithm. A algorithm plans the initial path of the mechanical claw from the current position to the target garbage position, and genetic algorithm optimizes the path to avoid collision and improve efficiency. Meanwhile, model predictive control (MPC) and force feedback control algorithm are adopted. MPC predicts the next action according to the mechanical claw motion model, and force feedback control algorithm ensures that the grasping force of the mechanical claw is appropriate, so as to avoid damaging the garbage or causing grasping failure. 5.The visual recognition based garbage sorting device and system according to claim 1, characterized in that: The garbage compression module drives the first push plate (6) and the second push plate (8) to move towards each other after the separate sorting is completed, compresses the remaining garbage, reduces the volume of the garbage, and improves the storage efficiency. The garbage compression module uses the combination of finite element analysis algorithm and fuzzy logic algorithm. Finite element analysis algorithm simulates the stress deformation of garbage in the compression process to determine the optimal compression path and force range. Fuzzy logic algorithm adjusts the compression parameters according to the type and quantity of garbage to ensure the compression effect and avoid excessive pressure on the garbage treatment equipment. Meanwhile, the pressure control algorithm based on particle swarm optimization algorithm (PSO) and simulated annealing algorithm is used. PSO algorithm optimizes the compression pressure, and simulated annealing algorithm further adjusts to prevent local optimal solution, ensuring stable and efficient compression process. 6.The visual recognition based garbage sorting device and system according to claim 1, characterized in that: The cleaning module drives the brush (1104) to move on the surface of the holder (4) after the garbage sorting operation is completed, removes the residual plastic pieces and small garbage, keeps the holder (4) clean, and ensures that the subsequent sorting operation is not affected. The cleaning module uses the combination of watershed algorithm and region growing algorithm in image segmentation algorithm. Watershed algorithm coarsely segments the image on the surface of the holder, and region growing algorithm further accurately segments the residual garbage area to determine the cleaning target. The movement control of the brush uses the combination of ant colony algorithm and Dijkstra algorithm. Ant colony algorithm plans the general cleaning path of the brush, and Dijkstra algorithm optimizes the path to make the brush cover the whole holder surface with the shortest path, realizing efficient cleaning. 7.The visual recognition based garbage sorting device and system according to claim 1, characterized in that: The collection barrel control module controls the first side plate (14) and the second side plate (15) to rotate accurately through a motor, thereby controlling the exposure and storage of the first collection barrel (9) and the second collection barrel (10), and achieving the accurate storage of sorted garbage in the corresponding collection barrel. A partition plate is arranged in the first collection barrel (9) for classified collection of different types of garbage. The collection barrel control module uses a neural network prediction algorithm and a Kalman filtering algorithm. The neural network prediction algorithm predicts the usage of the collection barrel according to the garbage sorting progress and type, and the Kalman filtering algorithm filters and corrects the prediction result, thereby improving the prediction accuracy. The motor control adopts a combination of a sliding film variable structure control algorithm and a robust control algorithm. The sliding film variable structure control algorithm ensures the rapidity and accuracy of the rotation of the side plate, and the robust control algorithm enhances the anti-interference ability of the system to the changes in motor parameters and external interference, thereby ensuring the accurate control of the collection barrel. 8.The visual recognition based garbage sorting device and system according to claim 1, characterized in that: The parameter optimization module optimizes and adjusts the identification parameters of the visual identification module, the motion parameters of the mechanical claw driving module, the compression force parameters of the garbage compression module, the cleaning path parameters of the cleaning module, and the rotation angle parameters of the collection barrel control module according to the feedback information of multiple sorting operations, thereby improving the overall performance and sorting accuracy of the system. The parameter optimization module combines a multi-objective genetic algorithm (MOGA) and a biological evolution algorithm (BEA). The MOGA optimizes multiple parameters simultaneously and considers the balance between multiple objectives. The BEA simulates the biological evolution process to further search for a better parameter combination. Meanwhile, a local optimization algorithm based on a gradient descent algorithm and a trust region algorithm is combined. The gradient descent algorithm quickly finds the optimal solution direction in a local area, and the trust region algorithm ensures the stability and convergence of the optimization process, thereby avoiding falling into a local optimal solution.

Images