Intelligent classification garbage can, using method and identification model establishing method

Abstract

The invention belongs to the technical field of garbage classification, and discloses an intelligent classification garbage can, a using method and a garbage recognition model building method.The intelligent classification garbage can comprises a box shell and a processing unit, a display device is arranged outside the box shell, and a first conveying belt and a second conveying belt are arranged on the first side face; the second conveyor belt is arranged below the first conveyor belt; an image acquisition device is arranged above the first conveying belt; a push rod mechanism is arranged beside the second conveying belt; the pushing rod moves from the first side face to the second side face. The end part of the push rod is connected with a push plate; a baffle is arranged on the push rod and located between the push plate and the drive motor. A classification box is arranged at the bottom of the box shell. According to the garbage can, the classification requirement of various kinds of garbage can be met, the garbage can be prevented from falling off by mistake through the baffle, the residual treatment in the garbage can is monitored through ultrasonic ranging, and the condition in the garbage can can be monitored through the display device.

Description

Technical Field

[0001] This invention belongs to the field of waste sorting technology, specifically relating to an intelligent waste sorting bin, its usage method, and a method for establishing a recognition model. Background Technology

[0002] Against the backdrop of heightened environmental awareness and intelligent urban management, smart waste sorting technology has become a research hotspot. Traditional waste sorting relies on manual judgment, which is inefficient and prone to errors. While existing smart trash cans attempt to automate the process through sensors and image processing, they suffer from problems such as insufficient recognition accuracy, low efficiency, poor user experience, and limited scalability, making it difficult to meet practical application needs. Foreign products such as Canada's OSCAR system, Poland's Bin-e trash can, and South Korea's SuperBin, although exploring smart sorting earlier, suffer from issues such as limited sorting categories, high costs, or insufficient adaptability, making it difficult to meet domestic waste sorting standards and diverse scenario requirements.

[0003] Chinese patent publication number CN113371363A, entitled "Classified Trash Can, Intelligent Classified Trash Can Based on Deep Learning, and Classification Method," describes a patent application that utilizes a two-stage conveyor belt to automatically sort three types of trash into different bins, while also compressing recyclable waste. In one scenario, manual sorting and automatic bin placement can be achieved using control buttons; in another, automatic sorting and placement are performed based on machine vision and a deep learning model using convolutional neural networks. While this patent application uses visual recognition for classification, it is limited to only three types, making it unsuitable for various classification requirements, and it does not allow real-time monitoring of the remaining trash can's capacity. Summary of the Invention

[0004] To overcome the problems existing in the prior art, the present invention aims to provide an intelligent sorting trash can, a method of use, and a method for establishing a recognition model. It uses two conveyor belts and pushes trash into different compartments by a pusher plate through a pull-back mechanism. By controlling the conveyor belts, it can meet the sorting requirements of various types of trash and prevent trash from falling into the wrong place through baffles. It can monitor the remaining waste inside the trash can through ultrasonic ranging and monitor the situation inside the trash can through a display device. The processing unit realizes the full automation of the trash can.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides an intelligent sorting trash can, comprising: a casing and a processing unit. A display device is disposed outside the casing. The casing has a first side and a second side disposed opposite to each other inside. A first conveyor belt and a second conveyor belt are disposed on the first side. The second conveyor belt is disposed below the first conveyor belt. An image acquisition device is disposed above the first conveyor belt. A push rod mechanism is disposed beside the second conveyor belt. The push rod mechanism has a push rod and a drive motor, and the push rod moves from the first side to the second side. The drive motor of the push rod mechanism is disposed on the second side. A push plate is connected to the end of the push rod. A baffle is disposed on the push rod, and the baffle is located between the push plate and the drive motor. A sorting bin is disposed at the bottom of the casing. An ultrasonic ranging module is disposed above the sorting bin. The processing unit is communicatively connected to the display device, the image acquisition device, the first conveyor belt, the second conveyor belt, the push rod mechanism, and the ultrasonic ranging module. The processing unit is used for device control and trash identification.

[0006] Optionally, the length of the baffle is less than the length of the second conveyor belt, and the length of the pusher is greater than or equal to the length of the second conveyor belt; side baffles are vertically connected to both ends of the pusher, and the length of the side baffles is greater than the width of the second conveyor belt; the first conveyor belt is connected to a first conveyor belt drive motor; and the second conveyor belt is connected to a second conveyor belt drive motor.

[0007] Optionally, the sorting box has multiple parallel partitions inside, which divide the interior of the sorting box into multiple sorting areas.

[0008] Optionally, the top of the sorting box is provided with a sloping plate, the bottom end of which is connected to the sorting box, and the top end is located below the edge of the second conveyor belt near the second side.

[0009] Optionally, the sorting area of ​​the sorting bin includes a compressible waste area, and a compression through hole is provided on the side of the sorting bin facing the first side; a compression mechanism is provided next to the sorting bin, the compression mechanism has a compression rod facing the compression through hole, and the compression mechanism communicates with the processing unit.

[0010] Optionally, the processing unit is an ATmega328P microcontroller or an ATmega4809 microcontroller with an Arduino Nano series development board.

[0011] Secondly, the present invention provides a method for using the aforementioned intelligent sorting trash can, comprising the following steps: The garbage is put into the outer shell of the container and falls onto the first conveyor belt. The image acquisition device captures the image of the garbage and transmits the image data to the processing unit. The processing unit analyzes the images and identifies the type of waste; The processing unit controls the first conveyor belt to drop the waste into the second conveyor belt, and uses baffles to prevent the waste from falling off. The processing unit controls the second conveyor belt to transport the waste to the area above the corresponding sorting bin; The processing unit controls the pusher mechanism to move the pusher plate toward the second side, pushing the garbage off the second conveyor belt and into the sorting bin; the processing unit obtains the garbage storage information through the ultrasonic ranging module and displays it through the display device.

[0012] Thirdly, the present invention provides a method for establishing a waste identification model, which uses the processing unit of the aforementioned intelligent waste sorting bin to identify waste, including the following steps: Garbage image data is acquired through an image acquisition device as training data. The input garbage images are normalized and the image size is adjusted to a fixed size required by the method model. The backbone network consists of convolutional layers and pooling layers; the convolutional operation slides on the image through different convolutional kernels to extract edge and texture features in the image; the pooling layer is used to reduce the resolution of the feature map, reduce the amount of computation, and at the same time retain the main features. After features are extracted by the backbone network, the feature map is processed by the YOLOv5 detection head to predict the category and bounding box of the garbage target. For each predicted garbage target bounding box, calculate its intersection-union ratio (IU) with the ground truth bounding box; compare the IU calculation result with a set value. If the IU is greater than the set value, the prediction result for that location is considered correct; otherwise, the prediction result for that location is considered incorrect. If the location prediction result is correct, the predicted classification of the garbage target is determined. If the prediction is correct, it is determined as a correct identification; otherwise, it is determined as an incorrect identification. The garbage identification model is obtained by iterative training based on the judgment results.

[0013] Optionally, the CSPDarknet53 network architecture can be used to extract features from the preprocessed image.

[0014] Optionally, the intersection-union ratio of the bounding box and the ground truth box is the ratio of the intersection area of ​​the predicted bounding box and the ground truth box to the merged area of ​​the predicted bounding box and the ground truth box.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention provides an intelligent sorting trash can, its usage method, and a recognition model establishment method. Compared to existing pushing methods, it can deliver trash into different sorting areas by pulling back a pusher plate. By controlling the movement distance of the second conveyor belt, trash is delivered to designated areas. Both the pusher plate and the baffle in this invention can prevent trash from falling incorrectly onto the second conveyor belt. Existing dual-conveyor belt designs, due to the lack of properly designed baffles, are prone to trash falling incorrectly into sorting areas during the process of dropping trash from the upper conveyor belt to the lower conveyor belt, given the diverse types of trash. The baffle and pusher plate of this invention can both prevent trash from falling and provide temporary storage. Since both the pusher plate and the baffle plate are mounted on the push rod, they do not obstruct trash from falling to the next layer. Furthermore, this invention can monitor the remaining processing status of each category in the trash can in real time using an ultrasonic ranging module, thereby improving the space utilization and management efficiency of the trash can.

[0016] Furthermore, the present invention can compress compressible waste through a compression rod, thereby further improving space utilization efficiency.

[0017] Compared to traditional smart trash cans that rely on simple sensors, label scanning, or basic voice recognition, this invention's waste identification model establishment method employs the YOLOv5 algorithm. By using pooling layers to reduce feature map resolution and computational load, it retains key features, achieving high-precision real-time image recognition. Existing solutions are susceptible to environmental interference; changes in lighting affect label scanning, and noisy environments interfere with voice recognition. Furthermore, they struggle to accurately classify complex waste shapes and mixed waste. This invention, however, leverages the powerful deep learning capabilities of YOLOv5, processing feature maps through its detection head. This effectively addresses complex scenarios, improving recognition accuracy. Even in dimly lit underground parking lots or cluttered community corners, it can accurately identify waste categories. This invention significantly improves recognition accuracy and reduces false positives through its algorithm and hardware architecture. Utilizing the efficient processing power of the nano development board shortens recognition time and improves classification efficiency. The mechanical structure and algorithm possess excellent scalability, allowing software updates to adapt to different regional waste classification standards and diverse scenario requirements.

[0018] Furthermore, this invention employs the YOLOv5 algorithm and uses a nano development board as the processing unit, balancing recognition accuracy and equipment cost. The high-efficiency processing capability of the nano development board ensures real-time performance, significantly improving waste recognition efficiency and classification accuracy. Furthermore, the mechanical structure and algorithm of this invention's intelligent waste sorting bin possess good scalability. The nano development board facilitates algorithm updates, adapting to waste sorting standards in different regions and meeting diverse scenario requirements. The overall structural design of this invention is simple, and the localized operation of the nano development board reduces equipment costs, facilitating large-scale application in communities, shopping malls, transportation hubs, and other scenarios. Attached Figure Description

[0019] The accompanying drawings described herein are for illustrative purposes only and are not intended to limit the scope of the invention in any way. Furthermore, the shapes and proportions of the components in the drawings are merely schematic to aid in understanding the invention and do not specifically limit the shapes and proportions of the components. In the drawings: Figure 1 This is a perspective view of the intelligent sorting trash can according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the internal structure of the intelligent sorting trash can according to an embodiment of the present invention; Figure 3 This is a three-dimensional structural diagram of the intelligent sorting trash can according to an embodiment of the present invention; Figure 4 This is an internal front view of the intelligent sorting trash can according to an embodiment of the present invention.

[0020] The components include: 1. Dispensing port; 100. Box shell; 101. Top cover; 102. Slide plate; 2. Display device; 3. Push rod mechanism; 31. Baffle; 32. Push plate; 33. Side baffle; 4. Sorting box; 41. Inclined plate; 42. Divider plate; 43. Box flat plate; 5. Compression mechanism; 51. Compression through hole; 6. Ultrasonic ranging module; 71. First conveyor belt drive motor; 72. Second conveyor belt drive motor; 81. First conveyor belt; 82. Second conveyor belt; 9. Image acquisition device; 10. Processing unit. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0022] Therefore, the following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention.

[0023] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper", "lower", "horizontal", "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of the invention is usually placed during use, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0024] When an element is referred to as being "set on" another element, it can be directly on the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only embodiments. The use of the term "horizontal" does not imply that the component is required to be absolutely horizontal, but rather that it may be slightly tilted. "Horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but rather that it may be slightly tilted.

[0025] It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the specification of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the specification and appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0027] The present invention will now be described in detail with reference to the accompanying drawings.

[0028] An intelligent sorting trash can of the present invention includes: a housing 100 and a processing unit 10. A display device 2 is disposed outside the housing 100. The interior of the housing 100 has a first side and a second side arranged opposite to each other. A first conveyor belt 81 and a second conveyor belt 82 are disposed on the first side. The second conveyor belt 82 is disposed below the first conveyor belt 81. An image acquisition device 9 is disposed above the first conveyor belt 81. A push rod mechanism 3 is disposed beside the second conveyor belt 82. The push rod mechanism 3 has a push rod and a drive motor, and the direction of movement of the push rod is determined by the first... The second side faces the push rod mechanism 3; the drive motor of the push rod mechanism 3 is located on the second side; the end of the push rod is connected to a push plate 32; a baffle 31 is provided on the push rod, and the baffle 31 is located between the push plate 32 and the drive motor; a sorting box 4 is provided at the bottom of the outer shell 100; an ultrasonic ranging module 6 is provided above the sorting box 4; the processing unit 10 is communicatively connected to the display device 2, the image acquisition device 9, the first conveyor belt 81, the second conveyor belt 82, the push rod mechanism 3, and the ultrasonic ranging module 6, and the processing unit 10 is used for device control and waste identification.

[0029] The present invention provides a method for using an intelligent sorting trash can, comprising the following steps: The garbage is put into the outer shell 100 of the container and falls onto the first conveyor belt 81. The image acquisition device 9 captures the garbage image and transmits the image data to the processing unit 10. Processing unit 10 analyzes the image and identifies the type of waste; The processing unit 10 controls the first conveyor belt 81 to drop the garbage into the second conveyor belt 82, and the baffle 31 prevents the garbage from falling. Processing unit 10 controls the second conveyor belt 82 to transport the waste to the area above the corresponding sorting area of ​​the sorting bin 4; The processing unit 10 controls the push rod mechanism 3 to move the push plate 32 toward the second side direction, pushing the garbage off the second conveyor belt 82 and into the sorting bin 4; the processing unit 10 obtains the garbage storage information through the ultrasonic ranging module 6 and displays it through the display device 2.

[0030] The present invention provides a method for establishing a waste identification model, which uses the processing unit 10 of the aforementioned intelligent waste sorting bin to identify waste, and includes the following steps: The image acquisition device 9 acquires garbage image data as training data, normalizes the input garbage images, and adjusts the image size to the fixed size required by the method model. The backbone network consists of convolutional layers and pooling layers; the convolutional operation slides on the image through different convolutional kernels to extract edge and texture features in the image; the pooling layer is used to reduce the resolution of the feature map, reduce the amount of computation, and at the same time retain the main features. After features are extracted by the backbone network, the feature map is processed by the YOLOv5 detection head to predict the category and bounding box of the garbage target. For each predicted garbage target bounding box, calculate its intersection-union ratio (IU) with the ground truth bounding box; compare the IU calculation result with a set value. If the IU is greater than the set value, the prediction result for that location is considered correct; otherwise, the prediction result for that location is considered incorrect. If the location prediction result is correct, the predicted classification of the garbage target is determined. If the prediction is correct, it is determined as a correct identification; otherwise, it is determined as an incorrect identification. The garbage identification model is obtained by iterative training based on the judgment results.

[0031] This invention's intelligent sorting trash can uses a pull-back mechanism to direct trash into different sorting areas, while a baffle 31 prevents trash from falling into the wrong category. This invention can monitor the remaining processing status of each category in real time using an ultrasonic ranging module 6, thereby improving the trash can's space utilization and management efficiency. This invention employs the YOLOv5 algorithm, using pooling layers to reduce feature map resolution, thus reducing computational load while preserving key features, achieving high-precision real-time image recognition. Existing solutions are susceptible to environmental interference; changes in lighting affect label scanning, noisy environments interfere with voice recognition, and they struggle to accurately sort complex trash shapes and mixed trash.

[0032] Example 1 The present invention provides an intelligent sorting trash can, comprising: a housing 100, wherein a trash disposal opening 1 is provided on the top of the housing 100.

[0033] Optionally, in this embodiment, the waste disposal port 1 is a rectangular window.

[0034] A display device 2 is provided on the top of the outer casing 100. Specifically, the display device 2 is a display screen.

[0035] Optionally, in this embodiment, the outer shell 100 is a rectangular frame with a top cover 101.

[0036] Furthermore, in this embodiment, the display device 2 is embedded in the top cover 101 of the housing shell 100.

[0037] The interior of the housing 100 has a first side and a second side arranged opposite to each other; a first conveyor belt 81 and a second conveyor belt 82 are arranged on the first side. The second conveyor belt 82 is located below the first conveyor belt 81.

[0038] The bottom of the top cover 101 of the outer casing 100 is provided with a sliding plate 102. One end of the sliding plate 102 is located directly below the waste disposal port 1, and the other end is located on the side of the first conveyor belt 81. Specifically, the sliding plate 102 is fixed to the inner side of the outer casing 100.

[0039] An image acquisition device 9 is disposed above the first conveyor belt 81. Optionally, the image acquisition device 9 is a camera. Optionally, the image acquisition device 9 is connected to the inner side of the housing 100 via a bracket.

[0040] The image acquisition device 9 is used to capture images of garbage.

[0041] The first conveyor belt 81, the second conveyor belt 82, and the top cover 101 are parallel to each other.

[0042] The first conveyor belt 81 is connected to a first conveyor belt drive motor 71 on its side.

[0043] The bottom of the second conveyor belt 82 is connected to the second conveyor belt drive motor 72.

[0044] The end of the first conveyor belt 81 is located above the end of the second conveyor belt 82. The starting point of the second conveyor belt 82 is located directly below the end of the first conveyor belt 81, extending outward along the conveying direction of the first conveyor belt 81.

[0045] A push rod mechanism 3 is provided on the side of the second conveyor belt 82. The push rod mechanism 3 is fixed to the inner side of the housing shell 100.

[0046] The push rod mechanism 3 has a push rod, and the direction of movement of the push rod is from the first side to the second side. Specifically, the push rod is perpendicular to the second conveyor belt 82.

[0047] The push rod of the push rod mechanism 3 is driven by a motor.

[0048] The end of the push rod is vertically connected to a push plate 32, and the two ends of the push plate 32 are vertically connected to side baffles 33.

[0049] The push plate 32 is perpendicular to the second conveyor belt 82. A baffle 31 is provided on the push rod, and the baffle 31 is located between the push plate 32 and the drive motor. The drive motor is located on the second side.

[0050] Optionally, the baffle 31 is located at the top edge of the second conveyor belt 82 near the second side, the baffle 31 is perpendicular to the second conveyor belt 82, the edge of the baffle 31 is provided with a notch, and the push rod is connected to the notch of the baffle 31.

[0051] The length of the push plate 32 is greater than the length of the second conveyor belt 82. The length of the side baffle 33 is greater than the width of the second conveyor belt 82.

[0052] A sorting box 4 is provided at the bottom of the outer shell 100. The interior of the sorting box 4 has multiple parallel partitions 42, which divide the interior of the sorting box 4 into multiple sorting areas. A ramp 41 is provided at the top of the sorting box 4, with one ramp 41 at the top of each sorting area. The ramp 41 is positioned between two partitions 42. The bottom end of the ramp 41 connects between the two partitions 42, and the top end is positioned below the edge of the second conveyor belt 82 near the drive motor of the push rod mechanism 3.

[0053] The ends of the two partition plates 42 are connected by a box plate 43, which is perpendicular to the partition plates 42.

[0054] Several areas of the sorting bin 4 are compressible waste areas, and compression through holes 51 are provided on the flat plate 43 of the bin body where the compressible waste area is located.

[0055] A compression mechanism 5 is provided next to the sorting bin 4. The compression mechanism 5 has a compression rod facing the compression through hole 51, which is used to compress the garbage to save space.

[0056] An ultrasonic ranging module 6 is installed above the sorting box 4. The ultrasonic ranging module 6 emits ultrasonic waves downwards, and each area of ​​the sorting box 4 has an ultrasonic ranging module 6 above it to detect the remaining space in each storage area and display it in real time through the display device 2.

[0057] The outer casing 100 houses a processing unit 10. Specifically, the processing unit 10 is a microcontroller with a development board.

[0058] Optionally, the processing unit 10 is an ATmega328P microcontroller or an ATmega4809 microcontroller with an Arduino Nano series development board.

[0059] The processing unit 10 is communicatively connected to the display device 2, the image acquisition device 9, the first conveyor belt drive motor 71, the second conveyor belt drive motor 72, the push rod mechanism 3, the compression mechanism 5, and the ultrasonic ranging module 6.

[0060] The processing unit 10 has a storage system. The development board of the processing unit 10 integrates the YOLOv5 algorithm, which identifies waste through images acquired by the image acquisition device 9. It receives image data from the image acquisition device 9 in real time and stores the user-disposed waste sorting data in the storage system.

[0061] The stored waste sorting data is used to optimize the recognition accuracy of the YOLOv5 algorithm and to statistically analyze the quantity of different types of waste to assist in the formulation of recycling plans.

[0062] The user interface provides human-computer interaction functions, and users can view the classification results through the interface; the waste classification storage area includes storage areas for recyclable waste, kitchen waste, hazardous waste and other waste.

[0063] When using this invention, the following steps are included: The garbage is put into the disposal port 1 and slides to the first conveyor belt 81 via the slide plate 102. The garbage image is immediately captured by the image acquisition device 9 and the image data is transmitted to the processing unit 10.

[0064] The processing unit 10 analyzes the image using the YOLOv5 algorithm to identify the type of waste, and sends the classification results to the display device 2.

[0065] The processing unit 10 controls the first conveyor belt 81 to drop the garbage into the second conveyor belt 82 through the first conveyor belt drive motor 71. During this process, the baffle 31 can prevent the garbage from falling.

[0066] The processing unit 10 controls the second conveyor belt 82 to transport the waste to the area above the corresponding sorting area of ​​the sorting bin 4 via the second conveyor belt drive motor 72.

[0067] The processing unit 10 controls the push rod mechanism 3 to push the push plate 32 off the second conveyor belt 82, and the waste falls into the sorting bin 4 through the inclined plate 41.

[0068] The processing unit 10 acquires waste storage information through the ultrasonic ranging module 6 and displays it through the display device 2.

[0069] When the compressible waste area reaches the set size, the processing unit 10 controls the compression mechanism 5 to compress the waste.

[0070] Optionally, the display device 2 is connected to the development board of the processing unit 10 via a USB data cable. The display device 2 transmits the acquired image data to the development board in real time in the form of digital signals.

[0071] Optionally, the YOLOv5 algorithm of processing unit 10 includes the following steps: Treat the garbage identification task as an object detection problem.

[0072] Image preprocessing: The input garbage image is first normalized to adjust the image size to a fixed size required by the algorithm model. In this embodiment, 416×416 pixels are used. The image pixel values ​​are normalized to the range [0,1] using the following formula:

[0073] Where I(x,y) is the pixel value of the original image at the (x,y) coordinates, and Inorm(x,y) is the normalized pixel value. This step aims to give image data of different sizes and pixel value ranges a unified format, which is convenient for subsequent model processing.

[0074] Feature extraction: A backbone network consisting of convolutional layers and pooling layers is used. In this embodiment, CSPDarknet53 is used to extract features from the preprocessed image.

[0075] Convolution operations extract edge and texture features from an image by sliding different convolution kernels across it. The formula is as follows:

[0076] Where Fi,j is the feature value of the convolutional feature map at position (i,j), I is the input image, K is the convolution kernel, k is the kernel size, and b is the bias.

[0077] Pooling layers are used to reduce the resolution of feature maps, thereby reducing computational cost, while preserving key features.

[0078] Target detection and classification: After feature extraction by the backbone network, the feature map is processed by YOLOv5's unique detection head to predict the category, location, and bounding box of the garbage target.

[0079] The detection head employs a multi-scale detection strategy, effectively detecting trash targets of varying sizes. For each predicted bounding box, its intersection-union ratio (LOU) with the ground truth bounding boxes (the trash category and location boxes labeled in the training set) is calculated using the following formula:

[0080] Where A and B are the areas of the predicted bounding box and the ground truth bounding box, respectively. The accuracy of the predicted bounding box is judged based on the IOU value and a preset threshold, and the waste category is determined.

[0081] The nano development board stores the garbage sorting data (including garbage images, identification categories, and disposal times) submitted by users into a local database.

[0082] These data are used to optimize the recognition accuracy of the YOLOv5 algorithm by continuously adding new classification data to the training set and retraining the model so that it can better adapt to newly emerging garbage features.

[0083] Optionally, the processing unit 10 uses an NVIDIA Jetson Nano development board. It features powerful GPU acceleration capabilities, employs a quad-core ARM A57 CPU, and is equipped with 4GB of LPDDR4 memory, enabling efficient execution of the YOLOv5 algorithm for real-time garbage identification and processing.

[0084] Optionally, the image acquisition device 9 uses an H65HD camera, which has a 1080P high-definition resolution and a frame rate of 30fps, and can clearly capture details of junk images, providing high-quality image data for the YOLOv5 algorithm.

[0085] Optionally, the display device 2 adopts a 7-inch DF-Robot DFR0533 TFT LCD capacitive touch screen, which supports multi-touch, provides a clear display, and facilitates user operation and viewing of classification results and receiving placement instructions.

[0086] Optionally, both the first conveyor belt drive motor 71 and the second conveyor belt drive motor 72 are MS42DDC stepper motors.

[0087] The MS42DDC stepper motor has high control precision, enabling precise control of the start, stop, and running speed of the conveyor belt to meet the needs of waste sorting and conveying.

[0088] The compression mechanism 5 uses a BMXL type electric push rod, which is driven by a motor to extend and retract, compressing recyclable waste and improving the space utilization of the storage area.

[0089] The ultrasonic ranging module 6 is model HC-SR04, which can detect the height of garbage in each storage area in real time, calculate the remaining space, and transmit the data to the processing unit 10 for real-time display via the display device 2.

[0090] Example 2 In this embodiment, experimental samples are first selected using stratified sampling: the waste is stratified according to its category (recyclable waste, kitchen waste, hazardous waste, and other waste), and samples are randomly selected within each stratum. For each of the 10 specified categories of waste (paper cups, plastic bottles, potato chunks, radish chunks, carrot chunks, batteries, expired medicines, ceramic shards, bricks, and pebbles), samples are randomly selected.

[0091] Each sample must maintain a 3:3:4 ratio of different states (clean / contaminated, intact / damaged, different lighting conditions) to ensure sample diversity.

[0092] Sample preprocessing: Data augmentation of the images (including random rotation ±15°, scaling 0.8-1.2 times, and brightness adjustment ±20%).

[0093] Each original sample was labeled with a true bounding box and a category label, and the accuracy of the labeling was cross-validated by three waste sorting personnel.

[0094] Evaluation index calculation method: Accuracy calculation formula:

[0095] Where: TP (True Positive): The number of samples that are correctly identified as positive, such as plastic bottles being correctly identified as recyclable waste.

[0096] TN (True Negative): The number of samples that are correctly identified as negative (e.g., potato chunks are correctly identified as non-hazardous waste).

[0097] FP (False Positive): The number of negative samples that were incorrectly identified as positive samples (e.g., ceramic tiles were incorrectly identified as recyclable waste).

[0098] FN (False Negative): The number of positive samples that were incorrectly identified as negative samples (e.g., batteries were incorrectly identified as other waste).

[0099] The confusion matrix visually illustrates the identification overlap of each category of waste, as shown in Table 1 below: Table 1

[0100] Average precision (mAP) calculation: For each type of waste, calculate the precision-recall curve (PR curve) and calculate the area under the curve as the average precision (AP) for that type. Finally, take the average AP of all types as the overall evaluation index (mAP):

[0101]

[0102] in: P(Precision):

[0103] R (Recall):

[0104] N represents the total number of waste categories.

[0105] Controlled experimental design: Control variables: Keep the hardware parameters of the nano development board fixed, and only change the state parameters of the input image (lighting, angle, degree of contamination). Each experiment was repeated three times, and the average value was taken as the final result to reduce random error. The algorithm optimization iterative method uses 5-fold cross-validation: the dataset is divided into 5 parts, and 4 parts are used as the training set and 1 part as the test set in turn. The accuracy on the test set is calculated after each iteration, and training is stopped when the accuracy improves by less than 0.5% for three consecutive iterations.

[0106] Timeliness verification: Record the total time taken from acquisition to output classification results for a single image (including image transmission, preprocessing, feature extraction, and classification judgment) for 1000 consecutive tests, calculate the average time and standard deviation, and verify the stability of the algorithm.

[0107] In this embodiment, the Scikit-learn library of Python is used to calculate the confusion matrix and accuracy. In this embodiment, Matplotlib is used to plot the precision-recall curve. In this embodiment, outliers (data that exceed the mean ± 3 times the standard deviation) are removed during the statistical process.

[0108] By using the above standardized statistical methods, we can ensure the objectivity, repeatability, and statistical significance of the experimental results, which can accurately reflect the actual recognition effect of the YOLOv5 algorithm in smart waste sorting bins.

[0109] Recognition Accuracy: Statistics show that this smart sorting trash can achieves an accuracy rate of over 95% in recognizing common types of waste. For example, the accuracy rate for plastic bottles is 98%, aluminum cans 97%, and waste paper 96%. In cases of misidentification, the main causes are uniquely shaped or mixed-material waste, such as some paper packaging boxes with metal decorations being mistakenly identified as metals in recyclable waste. However, as new classification data is continuously added to the local database to optimize and train the YOLOv5 algorithm, the recognition accuracy rate shows a gradual upward trend.

[0110] Space utilization: The use of the recyclable waste compactor increased the space utilization of the recyclable waste storage area by approximately 30%. During the testing period, the number of times the recyclable waste storage area overflowed decreased by 40% compared to when the compactor was not used, reducing the frequency of waste collection and improving the efficiency of the waste bins.

[0111] This implementation case fully demonstrates that the intelligent sorting trash can of the present invention has a high recognition accuracy, a good user experience and excellent space utilization in practical applications. It can effectively solve the problems existing in the current intelligent trash cans and provide an efficient and reliable solution for waste sorting.

[0112] A total of 10,000 image samples were selected from four common types of waste (recyclable waste, kitchen waste, hazardous waste, and other waste), with 2,500 images for each type. These images cover different lighting conditions (strong light / weak light / backlight), angles (front / side / tilted), and pollution levels (clean / stained / damaged), as shown in Table 2 below.

[0113] Table 2

[0114] Confidence threshold: 0.5 (predicted bounding box confidence ≥ 0.5 is considered valid recognition); IoU threshold: 0.5 (predicted bounding box IoU ≥ 0.5 with ground truth box is considered a correct match); Number of training iterations: 300; Learning rate: 0.001.

[0115] Predict the confidence level of the waste within the bounding box to belong to a certain category (e.g., "98% probability of belonging to recyclable waste" or "1% probability of belonging to other waste"), and finally select the category with the highest probability as the "category label" of the prediction box.

[0116] If IoU ≥ preset threshold (0.5 in the document), it means that the predicted bounding box and the ground truth bounding box are highly overlapping, and we can proceed to the next step of category verification. If IoU < 0.5, it is directly judged as a location prediction error, and no further category judgment is needed (e.g., if the bounding box of "ceramic fragment" is predicted to be the location of "plastic bottle", even if the category is guessed correctly, it is still considered an error).

[0117] Only after the IoU meets the threshold are the class labels of the predicted bounding boxes compared with the class labels of the ground truth bounding boxes. If both are consistent (e.g., the predicted box is labeled "recyclable waste" and the actual box is also labeled "recyclable waste"), it will be considered a correct identification. If they are inconsistent (e.g., the predicted box is labeled "recyclable waste" and the actual box is labeled "other waste"), even if the IoU is high, it will still be considered an "incorrect identification" (the location is correct but the category is wrong).

[0118] Image preprocessing: In this embodiment, a plastic bottle image (pixel value range 0-255) is normalized as follows: Original pixel value: I(100,200)=150 (pixel value at coordinates (100,200) in the image).

[0119] After normalization:

[0120] Feature extraction uses a 3×3 convolution kernel to perform convolution calculations on the preprocessed image: convolution kernel Bias b=0 Local region of the input image:

[0121] Convolution result: Fi,j=(0.2×1+0.3×0+0.4×(-1))+(0.5×1+0.6×0+0.7×(-1))+(0.8×1+ 0.9×0+1.0×(-1))+0 =(0.2-0.4)+(0.5-0.7)+(0.8-1.0) =(-0.2)+(-0.2)+(-0.2)=-0.6 Object detection and classification (IoU calculation) Predicted bounding box A (plastic bottle): coordinates (50,80) to (150,200), area = 100 × 120 = 12000 Real frame B (plastic bottle): coordinates (55, 85) to (145, 195), area = 90 × 110 = 9900 Intersection region: coordinates (55, 85) to (145, 195), area = 90 × 110 = 9900 Union region: 12000 + 9900 - 9900 = 12000 IoU = 9900 / 12000 = 0.825 (≥0.5, considered correct recognition) The classification accuracy is shown in Table 3 below: Table 3

[0122] The recognition results in different scenarios are shown in Table 4 below: Table 4

[0123] The comparison of the algorithm before and after optimization is shown in Table 5 below: Table 5

[0124] Results of the analysis: Recognition accuracy: The YOLOv5 algorithm achieved an average accuracy of 96.0% on the test set, with the highest accuracy (98.0%) for recyclable waste, mainly due to its stable shape characteristics; the accuracy of hazardous waste and other waste was relatively lower due to the similarity of some sample characteristics (such as small batteries and ceramic fragments), but it still remained above 94%.

[0125] Environmental adaptability: In complex scenarios such as low light, tilt, and pollution, the algorithm can still maintain an accuracy of over 90%, which is better than traditional recognition methods based on simple feature matching (the accuracy in complex scenarios is usually less than 70%).

[0126] Optimization results: By continuously adding new classification data (such as new types of packaging waste), the algorithm accuracy improved by 5.5%, proving that the iterative optimization mechanism of the local database is effective.

[0127] The data above demonstrates that the YOLOv5 algorithm of this invention, combined with the hardware architecture of the nano development board, can meet the high precision and real-time requirements of intelligent waste sorting, and is significantly superior to existing recognition schemes.

[0128] This invention employs the YOLOv5 algorithm, which is trained on a large amount of garbage image data to construct a deep convolutional neural network model. During identification, the model extracts and matches features from garbage images captured by a high-definition camera to accurately determine the garbage category. For common recyclables such as plastic bottles and aluminum cans, it can accurately identify their material and shape characteristics, compare them with samples in the training set, and determine the classification.

[0129] This invention uses a nano development board as the processing unit, which possesses powerful computing capabilities and low power consumption. The nano development board can quickly run the YOLOv5 algorithm to process image data in real time, with a very short time from image acquisition to outputting classification results, meeting the rapid classification requirements in practical applications.

[0130] This invention uses a display device 2 as the user interface, which displays the sorting results in real time when the user disposes of garbage. If the garbage is disposed of incorrectly, the correct sorting information and instructions will pop up immediately.

[0131] The intelligent sorting trash can of this invention adopts a modular design in its mechanical structure, making it easy to adjust or replace components according to different needs. In terms of algorithms, the nano development board facilitates software updates, allowing for optimized training of the YOLOv5 algorithm model based on new trash sorting standards or trash types, adapting it to diverse scenarios.

[0132] Unless otherwise specified, the equipment components involved in the above embodiments are all conventional equipment components, and the structural settings, working methods or control methods involved are all conventional settings, working methods or control methods in the art unless otherwise specified.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solutions of the present invention, as long as they do not depart from the spirit and scope of the technical solutions of the present invention, should be covered within the scope of the claims of the present invention.

Claims

1. A smart sorting trash can, characterized in that, include: The enclosure (100) and processing unit (10) are provided. A display device (2) is provided outside the enclosure (100). The enclosure (100) has a first side and a second side arranged opposite to each other inside. A first conveyor belt (81) and a second conveyor belt (82) are provided at the first side. The second conveyor belt (82) is located below the first conveyor belt (81). An image acquisition device (9) is provided above the first conveyor belt (81). A push rod mechanism (3) is provided next to the second conveyor belt (82). The push rod mechanism (3) has a push rod and a drive motor. The direction of movement of the push rod is from the first side to the second side. The drive motor of the lever mechanism (3) is located on the second side; the end of the push rod is connected to a push plate (32); a baffle (31) is provided on the push rod, and the baffle (31) is located between the push plate (32) and the drive motor; a sorting box (4) is provided at the bottom of the outer shell (100); an ultrasonic ranging module (6) is provided above the sorting box (4); the processing unit (10) is communicatively connected to the display device (2), the image acquisition device (9), the first conveyor belt (81), the second conveyor belt (82), the push rod mechanism (3) and the ultrasonic ranging module (6), and the processing unit (10) is used for device control and garbage identification.

2. The intelligent sorting trash can according to claim 1, characterized in that, The length of the baffle (31) is less than the length of the second conveyor belt (82), and the length of the push plate (32) is greater than or equal to the length of the second conveyor belt (82); the two ends of the push plate (32) are vertically connected to side baffles (33), and the length of the side baffles (33) is greater than the width of the second conveyor belt (82); the first conveyor belt (81) is connected to a first conveyor belt drive motor (71); and the second conveyor belt (82) is connected to a second conveyor belt drive motor (72).

3. The intelligent sorting trash can according to claim 1, characterized in that, The sorting box (4) has multiple parallel partitions (42) inside, which divide the interior of the sorting box (4) into multiple sorting areas.

4. The intelligent sorting trash can according to claim 3, characterized in that, The top of the sorting box (4) is provided with a sloping plate (41), the bottom end of which is connected to the sorting box (4), and the top end is located below the edge of the second conveyor belt (82) near the second side.

5. The intelligent sorting trash can according to claim 3, characterized in that, The sorting area of ​​the sorting bin (4) includes a compressible waste area. The sorting bin (4) of the compressible waste area has a compression through hole (51) on one side facing the first side. A compression mechanism (5) is provided next to the sorting bin (4). The compression mechanism (5) has a compression rod facing the compression through hole (51). The compression mechanism (5) communicates with the processing unit (10).

6. The intelligent sorting trash can according to claim 1, characterized in that, The processing unit (10) is an ATmega328P microcontroller or an ATmega4809 microcontroller with an Arduino Nano series development board.

7. A method of using an intelligent sorting trash can according to any one of claims 1 to 6, characterized in that, Includes the following steps: The garbage is put into the outer shell (100) of the container and falls onto the first conveyor belt (81). The garbage image is captured by the image acquisition device (9) and the image data is transmitted to the processing unit (10). The processing unit (10) analyzes the image and identifies the type of waste; The processing unit (10) controls the first conveyor belt (81) to drop the garbage into the second conveyor belt (82), and prevents the garbage from falling through the baffle (31); The processing unit (10) controls the second conveyor belt (82) to transport the waste to the area above the corresponding sorting area of ​​the sorting bin (4); The processing unit (10) controls the push rod mechanism (3) to make the push plate (32) move toward the second side direction, pushing the garbage off the second conveyor belt (82) and into the sorting bin (4); the processing unit (10) obtains the garbage storage information through the ultrasonic ranging module (6) and displays it through the display device (2).

8. A method for establishing a waste identification model, characterized in that, The processing unit (10) of the intelligent sorting trash can according to any one of claims 1 to 6 identifies the trash, including the following steps: The garbage image data is acquired by the image acquisition device (9) as training data, and the input garbage image is normalized to adjust the image size to the fixed size required by the method model. The backbone network consists of convolutional layers and pooling layers; the convolutional operation slides on the image through different convolutional kernels to extract edge and texture features in the image; the pooling layer is used to reduce the resolution of the feature map, reduce the amount of computation, and at the same time retain the main features. After features are extracted by the backbone network, the feature map is processed by the YOLOv5 detection head to predict the category and bounding box of the garbage target. For each predicted garbage target bounding box, calculate its intersection-union ratio (IU) with the ground truth bounding box; compare the IU calculation result with a set value. If the IU is greater than the set value, the prediction result for that location is considered correct; otherwise, the prediction result for that location is considered incorrect. If the location prediction result is correct, the predicted classification of the garbage target is determined. If the prediction is correct, it is determined as a correct identification; otherwise, it is determined as an incorrect identification. The garbage identification model is obtained by iterative training based on the judgment results.

9. The method for establishing a waste identification model according to claim 8, characterized in that, The CSPDarknet53 network architecture is used to extract features from the preprocessed image.

10. A method for establishing a garbage identification model according to claim 8, characterized in that, The intersection-union ratio of the predicted bounding box and the ground truth box is the ratio of the intersection area of ​​the predicted bounding box and the ground truth box to the merged area of ​​the predicted bounding box and the ground truth box.

Images