Intelligent classification dustbin based on image recognition sorting

Abstract

The invention discloses an intelligent classification dustbin based on image recognition sorting, and belongs to the field of intelligent garbage classification equipment, the intelligent classification dustbin is characterized in that a garbage receiving box is arranged at the upper part in a main box body, a plurality of garbage storage boxes are arranged at the lower part in the main box body, and an outlet of the garbage receiving box is connected with each garbage storage box through a sorting mechanism; a garbage receiving opening is formed in the side surface of the garbage receiving box of the main box body; the visual identification and classification device is arranged in the main box body and can collect and receive garbage images of garbage from the garbage receiving port, perform feature extraction and classification and identification on the garbage images by using a convolutional neural network, determine garbage categories and send sorting control signals to the sorting mechanism according to the garbage categories; and the sorting mechanism can sort the received garbage into the corresponding garbage storage boxes according to the corresponding sorting control signals sent by the visual identification and classification device. According to the garbage can, the garbage classification accuracy and the operation efficiency can be improved, and garbage classification is promoted to develop towards the intelligent, green and sustainable direction.

Description

Technical Field

[0001] This invention relates to the field of intelligent waste sorting equipment, and in particular to an intelligent waste sorting bin based on image recognition sorting. Background Technology

[0002] In recent years, people have paid increasing attention to environmental protection, leading to growing pressure on urban waste management. Traditional trash cans only collect waste and cannot effectively sort or recycle it. With the development of smart technology, smart waste sorting bins have gradually emerged, capable of sorting, compressing, and recycling. However, most products on the market are currently high-cost, high-end equipment, and their widespread adoption is insufficient.

[0003] my country's waste sorting and intelligent environmental protection are rapidly advancing, creating a strong demand for intelligent identification and sorting technologies among related enterprises. Sanitation equipment companies can leverage convolutional neural networks to improve waste identification efficiency; smart city integrators can use Raspberry Pi and the Internet of Things (IoT) for real-time monitoring and data management; and community property management companies can reduce outdoor power consumption through solar power and energy-saving hardware. This invention's technology system can provide intelligent identification, automatic control, and energy-saving management support for various types of enterprises, promoting the intelligent, systematic, and low-carbon development of waste sorting.

[0004] Beijing WeiboYi Technology Co., Ltd. discloses a CLIP model training method, electronic device, and computer program product, which is a visual-based garbage classification solution. The method includes: acquiring video samples and their corresponding descriptive text; extracting fusion temporal features of the video using the visual encoder of the CLIP model, and extracting adapter enhancement features of the descriptive text using the text encoder; and jointly training the CLIP model by fusing the features of both to improve the accuracy of video-text matching and model performance. This solution only studies visual information extraction and does not cover classification domains for different categories.

[0005] Zhejiang Sci-Tech University has disclosed a real-time yarn bobbin monitoring method based on a coupled algorithm of machine vision and deep learning, which is a solution combining machine vision and neural networks. This method acquires RGB and depth images of the yarn bobbin at different pitch angles using a laser and an RGB-D camera, extracting the light spot and the center position and contour information of the yarn bobbin. It then combines computer vision and deep learning algorithms to calculate the yarn bobbin edge distance, and fuses and averages the results, significantly reducing measurement errors and achieving high-precision and stable detection of the yarn bobbin edge distance. This research focuses on deep learning and does not involve the field of machine vision for recognizing physical objects.

[0006] The promotion of waste sorting and recycling technologies in my country still faces many technical bottlenecks and practical problems.

[0007] Existing waste sorting technologies generally suffer from problems such as reliance on manual labor, low automation, high cost, insufficient identification accuracy, and low energy efficiency, which limit their promotion and application.

[0008] In view of this, the present invention is hereby proposed. Summary of the Invention

[0009] The purpose of this invention is to provide an intelligent sorting trash can based on image recognition, which can improve the accuracy and efficiency of trash sorting at a lower cost, promote the development of trash sorting towards intelligence, greening and sustainability, and thus solve the above-mentioned technical problems existing in the prior art.

[0010] The objective of this invention is achieved through the following technical solution: A smart sorting trash can based on image recognition includes: The main container, waste receiving bins, multiple waste collection bins, sorting mechanism, and visual recognition classification device; among them, The upper part of the main container is equipped with a garbage receiving box, and the lower part is equipped with multiple garbage collection boxes. The outlet of the garbage receiving box is connected to each garbage collection box through a sorting mechanism. The garbage receiving box of the aforementioned main body is provided with a garbage receiving port on its side; The visual recognition and classification device is installed inside the main container. This device can collect images of the received waste from the waste receiving port, and use a convolutional neural network to extract features and classify the waste images to determine the category of the received waste. Based on the category of the received waste, it sends a corresponding sorting control signal to the sorting mechanism. The sorting mechanism is electrically connected to the visual recognition and sorting device, and can sort the received garbage into the corresponding garbage collection bin according to the corresponding sorting control signal sent by the visual recognition and sorting device.

[0011] Compared with existing technologies, the intelligent sorting trash can based on image recognition provided by this invention has the following advantages: By incorporating a controllable sorting mechanism connecting a waste receiving bin and multiple waste collection bins within a main bin, and utilizing a visual recognition and classification device based on a convolutional neural network to identify waste images, this intelligent waste sorting bin controls the sorting mechanism to dispose of waste into the corresponding collection bins based on the identified waste type. This system achieves accurate waste sorting after accurately identifying the waste type. The integrated hardware and software design improves the accuracy and operational efficiency of waste sorting. Simultaneously, it achieves high stability and practicality at a lower cost, effectively addressing the pain points of traditional waste bins and promoting the intelligent, green, and sustainable development of waste sorting. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of the overall structure of an intelligent sorting trash can based on image recognition, provided in an embodiment of the present invention.

[0014] Figure 2 This is a top view of the intelligent sorting trash can based on image recognition, provided in an embodiment of the present invention.

[0015] Figure 3 This is a schematic diagram of the internal structure of an intelligent sorting trash can based on image recognition, provided in an embodiment of the present invention.

[0016] Figure 4 This is a schematic diagram of the inner bottom structure of an intelligent sorting trash can based on image recognition, provided in an embodiment of the present invention.

[0017] Figure 5 This is a side view of the intelligent sorting trash can based on image recognition, provided in an embodiment of the present invention.

[0018] Figure 6 This is a flowchart illustrating the control mechanism of an intelligent sorting trash can based on image recognition, provided as an embodiment of the present invention.

[0019] Figure 7 This is a schematic diagram of the convolutional neural network structure of the visual recognition and sorting device for an intelligent sorting trash can based on image recognition, provided in an embodiment of the present invention.

[0020] Figure 8 This is a flowchart illustrating the processing flow of an intelligent sorting trash can based on image recognition, provided as an embodiment of the present invention.

[0021] The components marked in the diagram are: 1- Caster wheel; 2- Vertical baffle; 3- Lower push-pull baffle; 4- 57 servo stepper motor; 5- Dispensing port enclosure; 6- Upper perimeter baffle; 7- LCD display; 8- Lower perimeter baffle; 9- Horizontal baffle; 10- Dispensing port tray; 11- Transformer; 12- L298N speed controller; 13- Alarm light; 14- Mega2560 development board; 15- 57 servo stepper motor; 16- Battery; 17- K210 vision module; 18- Smoke alarm; 19- Emergency stop switch; 20- Meshing gear; 21- Meshing rack and pinion partition; 22- Ranging Bluetooth; 23- Tray bracket; 24- Tray baffle; 25- Tray; 26- Lower motor gasket; 27- Aluminum profile outer frame; 28- Lower support vertical bracket. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the specific content of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments, which do not constitute a limitation of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0023] First, the following explanations are provided for the terms that may be used in this article: The term "and / or" means that either or both can be achieved simultaneously. For example, X and / or Y means that it includes both "X" or "Y" as well as the three cases of "X and Y".

[0024] The terms "comprising," "including," "containing," "having," or other similar semantic descriptions should be interpreted as non-exclusive inclusion. For example, including a technical feature element (such as raw material, component, ingredient, carrier, dosage form, material, size, part, component, mechanism, device, step, process, method, reaction conditions, processing conditions, parameter, algorithm, signal, data, product or article of manufacture, etc.) should be interpreted as including not only the expressly listed technical feature element, but also other technical feature elements that are not expressly listed and are well-known in the art.

[0025] The term "composed of" excludes any technical features not expressly listed. When used in a claim, it closes the claim to exclude all technical features other than those expressly listed, except for associated conventional impurities. If the term appears only in a clause of a claim, it limits the claim to the elements expressly listed in that clause; elements recited in other clauses are not excluded from the overall claim.

[0026] Unless otherwise explicitly specified or limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this document according to the specific circumstances.

[0027] The terms “center,” “longitudinal,” “lateral,” “length,” “width,” “thickness,” “upper,” “lower,” “front,” “back,” “left,” “right,” “vertical,” “horizontal,” “top,” “bottom,” “inner,” “outer,” “clockwise,” and “counterclockwise” indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience and simplification of description and do not imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this document.

[0028] The solution provided by this invention will be described in detail below. Contents not described in detail in the embodiments of this invention are prior art known to those skilled in the art. Where specific conditions are not specified in the embodiments of this invention, they shall be performed according to conventional conditions in the art or conditions recommended by the manufacturer. Reagents or instruments used in the embodiments of this invention whose manufacturers are not specified are all conventional products that can be purchased commercially.

[0029] like Figures 1 to 7 As shown, an embodiment of the present invention provides an intelligent sorting trash can based on image recognition, comprising: The main container, waste receiving bins, multiple waste collection bins, sorting mechanism, and visual recognition classification device; among them, The upper part of the main container is equipped with a garbage receiving box, and the lower part is equipped with multiple garbage collection boxes. The outlet of the garbage receiving box is connected to each garbage collection box through a sorting mechanism. The garbage receiving box of the aforementioned main body is provided with a garbage receiving port on its side; The visual recognition and classification device is installed inside the main container. This device can collect images of the received waste from the waste receiving port, and use a convolutional neural network to extract features and classify the waste images to determine the category of the received waste. Based on the category of the received waste, it sends a corresponding sorting control signal to the sorting mechanism. The sorting mechanism is electrically connected to the visual recognition and sorting device, and can sort the received garbage into the corresponding garbage collection bin according to the corresponding sorting control signal sent by the visual recognition and sorting device.

[0030] Preferably, the above-mentioned intelligent sorting trash can further includes: a status monitoring device, a walking mechanism and a control device. The status monitoring device is installed in the main bin and can monitor the height of the trash in each trash collection bin and issue an alarm signal when the height reaches a predetermined height. The walking mechanism is located at the bottom of the main body and can support the movement of the main body; The control device is electrically connected to the sorting mechanism and can control the start and stop of the sorting mechanism according to the control signal.

[0031] Preferably, the status monitoring device of the above-mentioned intelligent sorting trash can includes: Ultrasonic devices, alarm lights, and smoke detectors; among them, The ultrasonic device is installed inside the main container and located inside each garbage collection bin. It can monitor the height of the garbage in each garbage collection bin through ultrasonic waves and emit an electrical signal after the height reaches a predetermined height. The alarm light and smoke detector are both electrically connected to the ultrasonic device and can emit alarm light and smoke alarm sound according to the electrical signal of the ultrasonic device.

[0032] Preferably, in the above-mentioned intelligent sorting trash cans, the plurality of trash collection bins are at least two of the following: Household waste collection bins, kitchen waste collection bins, hazardous waste collection bins, and other waste collection bins.

[0033] Preferably, in the above-mentioned intelligent sorting trash can, the visual recognition sorting device includes: Camera assembly, convolutional neural network, and control signal output module; among which The camera assembly is located at the garbage receiving port on the side of the garbage receiving bin of the main body, and can capture garbage images from the garbage receiving port; The convolutional neural network is communicatively connected to the camera component and can use the garbage images captured by the camera component as input to extract features and classify the garbage images to determine the category of the received garbage. The control signal output module is communicatively connected to the convolutional neural network and can send corresponding sorting control signals to the sorting mechanism according to the type of waste received as output by the convolutional neural network.

[0034] See Figure 8 Preferably, the above-mentioned convolutional neural network includes: The system consists of a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a fully connected layer, and a regression layer; among which, The first convolutional layer can take the garbage image captured by the camera as input, perform the first convolutional processing to extract low-level features to obtain an initial feature map; The first pooling layer, connected to the first pooling layer, can perform a first downsampling process on the initial feature map output by the first convolutional layer to obtain a first compressed feature map; The second convolutional layer, connected to the first pooling layer, can perform a second convolutional process on the first compressed feature map output by the first pooling layer to extract high-level semantic features and obtain an intermediate feature map. The second pooling layer, connected to the second convolutional layer, can perform a second downsampling process on the intermediate feature map output by the second convolutional layer to obtain an intermediate compressed feature map; The fully connected layer is connected to the second pooling layer and is used to flatten the second compressed feature map output by the second pooling layer, and to perform weighted fusion and comprehensive discrimination on the high-dimensional semantic features contained therein to obtain the classification discrimination feature vector or classification score vector corresponding to each waste category. The regression layer, connected to the fully connected layer, is used to perform normalized regression processing on the classification feature vector or classification score vector corresponding to each waste category output by the fully connected layer, output the probability distribution of each waste category, and finally determine the waste category based on the category with the largest probability value in the probability distribution.

[0035] Preferably, in the above-mentioned convolutional neural network, the first convolutional layer adopts a 3×3 standard convolutional structure; The first pooling layer adopts a 2×2 max pooling structure; The second convolutional layer employs a depth-separable convolution or dilated convolution structure; The second pooling layer adopts a 2×2 max pooling structure; The fully connected layer adopts a fully connected structure with ReLU activation; The regression layer uses a Softmax regression structure.

[0036] Preferably, the above-mentioned convolutional neural network further includes: The saliency enhancement unit is connected between the first convolutional layer and the first pooling layer. It can perform saliency enhancement on the main garbage region in the intermediate feature map output by the first convolutional layer to obtain an initial feature map with enhanced features of the main garbage region, and output the initial feature map to the first pooling layer.

[0037] And / or, the convolutional neural network further includes: The attention module, connected before or inside the fully connected layer, can enhance the attention of high-dimensional features of the second compressed feature map processed by the fully connected layer, thereby enhancing the feature selection capability of the fully connected layer.

[0038] Preferably, the above-mentioned saliency feature enhancement unit includes: The system includes a salient feature extraction module, a salient weight generation module, and a salient weighted fusion module; among which, The salient feature extraction module is connected to the first convolutional layer and can perform feature aggregation on the initial feature map output by the first convolutional layer in the spatial dimension to obtain a salient response map. The saliency weight generation module is connected to the saliency feature extraction module and can perform normalization and nonlinear mapping processing on the saliency response map obtained by the saliency feature extraction module to generate a saliency weight map. The saliency weighted fusion module is connected to the saliency weight generation module and the first convolutional layer respectively. It can expand the saliency weight map generated by the saliency weight generation module according to the channel dimension and multiply it element-wise with the initial feature map output by the first convolutional layer to obtain an initial feature map that enhances the features of the main garbage region. The attention module is either an SE channel attention module or a CBAM attention module.

[0039] The aforementioned convolutional neural network is a compact, computationally efficient, and feature-expressive convolutional neural network for garbage image classification. It can automatically identify garbage categories using garbage images captured by a camera as input. It connects the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the fully connected layer, and the regression layer in that order from front to back. The functions of each layer are as follows: (1) First convolutional layer (low-level feature extraction and saliency enhancement): The first convolutional layer is used to perform initial feature extraction on the input garbage image. It preferably adopts a 3×3 standard convolutional structure to extract low-level visual features such as edges and textures in the garbage image.

[0040] For a standard convolution with M input channels, K output channels, and N×N kernel size, the computational complexity satisfies the following formula: M×N 2 ×K×H×W; Where H×W is the spatial size of the output feature map.

[0041] Furthermore, a saliency enhancement unit can be integrated between the first convolutional layer and the first pooling layer to highlight the main features of the garbage image in the shallow stage of the network and suppress background interference information. By enhancing the regions with obvious visual response in the image, the network can generate a higher response to the main garbage region in the early stage of feature extraction, which is beneficial for distinguishing garbage targets in complex backgrounds, thereby improving the accuracy of subsequent feature extraction and classification.

[0042] The processing order for adding the saliency enhancement unit is as follows: input garbage image to the first convolutional layer to the saliency enhancement unit to the first pooling layer.

[0043] The saliency enhancement unit consists of a sequentially connected saliency feature extraction module, a saliency weight generation module, and a saliency weighted fusion module. The processing flow for saliency feature enhancement is as follows: (1) Saliency feature extraction is performed by the saliency feature extraction module: Let the initial feature map output by the first convolutional layer be... Where C is the number of channels and H×W is the spatial size.

[0044] The saliency feature extraction module performs feature aggregation on the initial feature map in the spatial dimension, which can be achieved using mean pooling or max pooling, to obtain the saliency response map: Where c represents the channel index, used to identify the sequence number of the feature map in the channel dimension, c=1,2,…,C; h represents the spatial position index of the feature map in the height direction, h=1,2,…,H; w represents the spatial position index of the feature map in the width direction, w=1,2,…,W; this saliency response map is used to reflect the importance of different spatial positions to the waste target.

[0045] (2) The significance weights are generated by the significance weight generation module: The obtained saliency response map S(h,w) is normalized and subjected to nonlinear mapping to generate a saliency weight map: A(h,w)=σ(S(h,w)); where σ(˙) represents the Sigmoid activation function, used to constrain the weight values ​​within the interval [0,1]. This saliency weight map A(h,w) maintains the same spatial dimension as the original feature map.

[0046] (3) Significance-weighted fusion is performed by the significance-weighted fusion module: The obtained saliency weight map A(h,w) is expanded along the channel dimension and multiplied element-wise with the initial feature map X(c,h,w) output by the first convolutional layer to obtain the saliency-enhanced feature map:

[0047] ; By using the above weighting method, the feature response corresponding to the main garbage area is enhanced, while the features of the background area are relatively suppressed.

[0048] (2) First pooling layer (downsampling and information retention): The first pooling layer is connected to the first convolutional layer, and preferably adopts a 2×2 max pooling structure to downsample the feature map output by the first convolutional layer.

[0049] By using max pooling, the spatial resolution of the feature map is reduced while retaining the maximum response value in the local region, thereby reducing the amount of computation and suppressing redundant information, providing a more compact input representation for subsequent deep feature extraction.

[0050] (3) Second convolutional layer (lightweighting and receptive field expansion): The second convolutional layer is connected to the first pooling layer to extract higher-level semantic features, preferably using at least one of depthwise separable convolutional structure or dilated convolutional structure.

[0051] (31) Depthwise separable convolution (reduces computational cost): In a preferred embodiment, the second convolutional layer adopts a depth-separable convolutional structure, and its computational complexity satisfies: N×N×K×H×W+M×K×H×W; By separating spatial convolution from channel convolution, the number of parameters and computational complexity are significantly reduced while maintaining feature representation capabilities, making it suitable for deployment on embedded or edge devices.

[0052] (32) Dential convolution (expanding the receptive field): In another preferred embodiment, the second convolutional layer employs a dilated convolution structure, and its convolution expression is as follows: ; Where d represents the hole ratio. This structure expands the effective receptive field of the network without increasing the number of parameters, which is beneficial for capturing large-volume garbage and long-range contextual information.

[0053] (4) Second pooling layer (feature compression and stability improvement): The second pooling layer is connected to the second convolutional layer, preferably using a 2×2 max pooling structure, to further compress the size of the feature map output by the second convolutional layer.

[0054] By setting up a two-level pooling structure, the network can maintain computational stability while abstracting features layer by layer, and reduce the parameter size of subsequent fully connected layers.

[0055] (5) Fully connected layer (attention-enhanced feature selection): The fully connected layer is connected to the second pooling layer and is used to comprehensively distinguish high-dimensional features. Preferably, a fully connected structure with a ReLU activation function is used.

[0056] More preferably, a channel attention (SE) module or a CBAM module is introduced before or inside the fully connected layer to enhance feature selection capabilities.

[0057] (51) SE channel attention mechanism: For the input feature map Performing global average pooling, we get: ; Channel weights are generated through a two-layer fully connected mapping: ; And recalibrate the original features: ; (52) CBAM attention mechanism: In another preferred approach, the CBAM module is used to first perform channel attention modeling on the features, and then perform spatial attention modeling, so as to achieve joint enhancement of key channels and key regions.

[0058] (6) Regression layer (classification output): The regression layer is connected to the fully connected layer, preferably using a Softmax regression structure, to output the probability distribution of the waste category and achieve the final waste category determination.

[0059] Through the organic synergy of the above layers, the first convolutional layer, combined with saliency enhancement, enables rapid localization of the main body of the waste; the second convolutional layer achieves both lightweight design and a large receptive field through depthwise separable convolution and dilated convolution; the fully connected layer, combined with SE or CBAM attention mechanisms, enhances feature discrimination capability; and the multi-level pooling structure effectively controls network complexity.

[0060] Preferably, to address issues such as class imbalance, label noise, and model convergence difficulties, this invention employs the following composite training loss and strategy to train the aforementioned convolutional neural network, including: (I) Design of Composite Loss Function: (11) Focal Loss (alleviating class imbalance and focusing on difficult cases): Using Focal Loss: This allows the convolutional neural network model to assign higher weights to difficult samples; when the weights of easily classified samples (Pt) are high, the weights decrease, and the convolutional neural network model learns more difficult examples (such as transparent plastic and mixed waste).

[0061] (12) Label Smoothing (alleviates overfitting): Smoothing the category label y: The target distribution is defined as follows: To reduce the risk of overfitting and improve the model's generalization ability.

[0062] (13) Hybrid Enhancement (Mixup / CutMix): Mixup improves stability through sample mixing and label interpolation; CutMix enhances the model's robustness to local information through region replacement.

[0063] (14) Category resampling + dynamic threshold optimization: Combining Focal loss, label smoothing and resampling strategies to improve the performance of long-tail categories.

[0064] (II) Distillation Training (Enhancing Generalization Ability): Using a teacher-student model structure, distillation loss is calculated using KL divergence: This allows for knowledge transfer and improves the classification accuracy of lightweight models.

[0065] (III) Quantitative Training (QAT): 8-bit quantization simulation is performed during the training phase to improve the efficiency of model deployment on mobile and edge devices.

[0066] (iv) Model structure pruning: By sparsifying the convolution kernel and using threshold selection and L1 / L2 regularization, unimportant weights are automatically discarded, thereby compressing the model size.

[0067] These improvements can reduce feature interference caused by lighting / angle / occlusion, improve the recognition rate of small targets (such as mineral water bottles and tiles), and achieve lower latency and longer battery life in Raspberry Pi / microcontroller environments with limited computing power, thus providing empirically proven improvements in accuracy and classification efficiency.

[0068] Preferably, in the above-mentioned intelligent sorting trash can, the sorting mechanism includes: Waste falling baffle, servo drive device, transmission device, rotating pallet mechanism, and control device; among which, The waste falling baffle is movably disposed between the lower outlet of the waste receiving box inside the main box and the multiple waste collection boxes below it; The drive shaft of the servo drive device is connected to the waste falling baffle through a transmission device, and can drive the waste falling baffle to open and close. The rotating pallet mechanism is located at the bottom of the main container. The rotating pallet mechanism has multiple mounting positions arranged in a ring, and each mounting position is equipped with a garbage collection bin. It can drive each garbage collection bin to rotate to the bottom of the garbage falling partition. The control device is electrically connected to the visual recognition and sorting device, the rotating pallet mechanism, and the servo drive device, respectively. It can control the servo drive device and the rotating pallet mechanism to drive the corresponding waste collection bin to rotate below the waste falling partition according to the corresponding sorting control signal sent by the visual recognition and sorting device.

[0069] In summary, the intelligent sorting trash can of this invention, by setting up a trash receiving bin and multiple trash storage bins connected by a controllable sorting mechanism in the main bin body, and by using a visual recognition and sorting device based on a convolutional neural network to recognize trash images, controls the sorting mechanism to put the trash into the corresponding trash storage bin according to the identified trash type, the intelligent sorting trash can can achieve accurate trash sorting after accurately identifying the trash type.

[0070] To more clearly demonstrate the technical solution and its effects provided by the present invention, the following detailed description of the solution provided by the embodiments of the present invention is provided with reference to specific examples.

[0071] Example 1 like Figures 1 to 6 As shown, this embodiment provides an intelligent sorting trash can based on image recognition. It employs a convolutional neural network (CNN) as the core algorithm, combined with machine vision technology, to achieve automatic identification and sorting of trash. The system acquires trash images through a Raspberry Pi camera, preprocesses them, and then inputs them into the CNN to extract features and perform classification. The classification results are controlled by a microcontroller, which drives a servo motor to drive the sorting mechanism, placing the trash into the corresponding category of trash bin. Simultaneously, the system incorporates an ultrasonic sensor module for full-load monitoring and utilizes its own battery and low-power circuitry to improve energy efficiency, creating a highly efficient, intelligent, and energy-saving integrated intelligent sorting trash can.

[0072] The intelligent sorting trash can in this embodiment includes: a main body, a trash receiving bin, multiple trash storage bins, a sorting mechanism, and a visual recognition sorting device; wherein... The upper part of the main container is equipped with a garbage receiving box, and the lower part is equipped with multiple garbage collection boxes. The outlet of the garbage receiving box is connected to each garbage collection box through a sorting mechanism. The garbage receiving box of the aforementioned main body is provided with a garbage receiving port on its side; The visual recognition and classification device is installed inside the main container. This device can collect images of the received waste from the waste receiving port, and use a convolutional neural network to extract features and classify the waste images to determine the category of the received waste. Based on the category of the received waste, it sends a corresponding sorting control signal to the sorting mechanism. The sorting mechanism is electrically connected to the visual recognition and sorting device, and can sort the received garbage into the corresponding garbage collection bin according to the corresponding sorting control signal sent by the visual recognition and sorting device.

[0073] The aforementioned intelligent sorting trash can utilizes a visual recognition sorting device. It uses a convolutional neural network to extract features and classify the collected trash images. Based on the recognition results, it determines the trash classification and controls the sorting mechanism to achieve automatic sorting and disposal of trash.

[0074] Furthermore, the aforementioned intelligent sorting trash can also include: a status monitoring device and a walking mechanism. The status monitoring device is installed inside the main bin and can monitor the height of the trash in each trash collection bin, and issue an alarm signal when the height reaches a predetermined height.

[0075] The walking mechanism is located at the bottom of the main body and can support the movement of the main body.

[0076] Preferably, the status monitoring device of the above-mentioned intelligent sorting trash can includes: an ultrasonic device, an alarm light, and a smoke alarm; wherein, The ultrasonic device is installed inside the main container and located inside each garbage collection bin. It can monitor the height of the garbage in each garbage collection bin through ultrasonic waves and emit an electrical signal after the height reaches a predetermined height. The alarm light and smoke detector are both electrically connected to the ultrasonic device and can emit alarm light and smoke alarm sound according to the electrical signal of the ultrasonic device.

[0077] By setting up status monitoring devices and combining ultrasonic sensing technology, the status of trash cans can be monitored. Specifically, ultrasonic sensors can be installed at the lower part of the middle baffle of the trash can and at the upper part of each corresponding compartment. Using the ultrasonic ranging function, when the trash inside any compartment exceeds a certain height, the ultrasonic device will transmit a signal to the display screen, thereby notifying sanitation workers to clean the trash inside the compartment in time. This achieves the design goals of high-precision sorting, low-energy operation, and automated management.

[0078] This embodiment provides an intelligent sorting trash can based on image recognition, specifically including: a visual recognition sorting device (including a vision mechanism, a 3D-printed vision support, and an aluminum profile support), a sorting mechanism (including a servo stepper motor, gears, and rack and pinion baffles), a status monitoring device (including an ultrasonic device, alarm lights, and smoke detectors), a control module (including an emergency stop switch, speed controller, and transformer), and an integrated walking mechanism (casters). The trash disposal process is as follows: the ultrasonic module detects the entry of trash, the visual recognition sorting device detects the type of trash being disposed of, the upper motor controls the movement of the lower gear and rack, and the lower motor drives the rotation of the tray.

[0079] The aforementioned sorting mechanism comprises: a waste falling partition, a servo drive device, a transmission device, a rotating pallet mechanism, and a control device; wherein, The waste falling baffle is movably disposed between the lower outlet of the waste receiving box inside the main box and the multiple waste collection boxes below it; The drive shaft of the servo drive device is connected to the waste falling baffle through a transmission device, and can drive the waste falling baffle to open and close. The rotating pallet mechanism is located at the bottom of the main container. The rotating pallet mechanism has multiple mounting positions arranged in a ring, and each mounting position is equipped with a garbage collection bin. It can drive each garbage collection bin to rotate to the bottom of the garbage falling partition.

[0080] The control device is electrically connected to the visual recognition and sorting device, the rotating pallet mechanism, and the servo drive device, respectively. It can control the servo drive device and the rotating pallet mechanism to drive the corresponding waste collection bin to rotate below the waste falling partition according to the corresponding sorting control signal sent by the visual recognition and sorting device.

[0081] This can be achieved by controlling a servo motor using a microcontroller. For example, instructions can be written in an Arduino and imported into the microcontroller to control the servo motor's movement. The specific hardware control process is as follows: Figure 7 As shown, a Raspberry Pi is used to coordinate and control the servo motor. By configuring the dataset and uploading the training library in the YOLOv3 environment, visual codes for recognizing different types of objects are obtained; then, a microcontroller is used to coordinate and control the movement of the servo motor.

[0082] The sorting mechanism consists of two layers. The upper layer uses a gear-rack structure as the transmission device, with a microcontroller controlling a servo motor to drive the gears to mesh with the rack, enabling the waste-falling partition to open and close linearly. The lower layer is equipped with a tray as a receiving device. The tray servo motor rotates to the corresponding waste collection bin (including: household waste collection bin, kitchen waste collection bin, hazardous waste collection bin, other waste collection bin, etc.) based on feedback signals from the visual recognition and sorting device, achieving accurate disposal of different types of waste.

[0083] In the intelligent sorting trash can of this invention, Figure 2 The presentation showcases a top-down view of an intelligent sorting trash can based on image recognition. The visual support structure comprises a 3D-printed tray and components machined using a Xingtai cutting machine, with the upper support surface made from a single laser-cut acrylic sheet. The front features a display screen for showing environmental information and includes an emergency stop button for safety protection; the sides are equipped with alarm lights and smoke detectors for safety monitoring and protection device operation.

[0084] In this invention, Figure 3The design showcases a structure based on a gear and rack meshing drive for an intermediate baffle. This module employs a split design, with the gears and rack modeled in 3D software and then laser-cut. This structure provides cushioning during waste descent and effectively separates the visual recognition time from the lower pallet motor's operation, improving system safety, sorting accuracy, and operational efficiency.

[0085] In the intelligent sorting trash can of this invention, Figure 4 The image shows the upward-facing structure of the smart waste sorting bin. Four omnidirectional wheels are installed at the bottom of the bin for easy handling and to reduce the workload of sanitation workers. An ultrasonic module is installed in the middle partition of the bin to detect when it is full by measuring distance, thus improving the efficiency of use and management.

[0086] In the intelligent sorting trash can of this invention, Figure 5 The lower tray structure of the intelligent waste sorting bin was showcased. The tray module consists of a drive motor, a ramp support, the tray body, and an upper baffle. The ramp support was 3D printed after 3D modeling, while the tray body and baffle were laser-cut. This structure efficiently guides waste deposited from the upper layer into the corresponding collection bins, serving a bridging and diversion function.

[0087] In the intelligent sorting trash can of this invention, Figure 6 The side view of the lower support structure for the motor and tray is shown. The support is composed of 3D-printed motor pads and laser-cut mortar. The motor pads are designed to a 1:1 scale with the bottom dimensions to cushion the load and extend the lifespan; the mortar structure isolates the tray from the waste at the bottom, preventing damage to the motor.

[0088] The aforementioned visual recognition and classification device is a module that uses a convolutional neural network (CNN) as its core algorithm to identify garbage images. The structure diagram of a CNN is shown below. Figure 7 As shown, it can maintain the original dimensions of the image and retain the original information of the image.

[0089] The processing flow of the image recognition-based intelligent sorting trash can in this embodiment is as follows: Figure 8 .

[0090] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0091] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims. The information disclosed in the background section is intended only to enhance the understanding of the overall background technology of the present invention and should not be construed as an admission or implication in any way that such information constitutes prior art known to those skilled in the art.

Claims

1. An intelligent classification garbage can based on image recognition sorting, characterized in that, The utility model relates to a garbage sorting device, including: The total box body, garbage receiving box, a plurality of garbage storage box, sorting mechanism and visual identification classification device, wherein, The upper portion in the total box body is provided with garbage receiving box, and the lower portion is provided with a plurality of garbage storage box, and the outlet of garbage receiving box is connected with each garbage storage box through sorting mechanism, The garbage receiving box side of the total box body is equipped with garbage receiving mouth, The visual identification classification device is arranged in the total box body, the visual identification classification device can collect the garbage image of received garbage from the garbage receiving mouth, and the garbage image is extracted and classified by using convolutional neural network, the category of received garbage is determined, and the corresponding sorting control signal is sent to the sorting mechanism according to the category of received garbage, The sorting mechanism is electrically connected with the visual identification classification device, and the received garbage can be sorted into the corresponding garbage storage box according to the corresponding sorting control signal sent by the visual identification classification device.

2. The smart sorting bin based on image recognition sorting according to claim 1, characterized in that, It also includes a state monitoring device, a walking mechanism and a control device, the state monitoring device is arranged in the total box body, can monitor the height of garbage in each garbage storage box, and sends an alarm signal when the height reaches a predetermined height, The walking mechanism is arranged at the bottom of the total box body and can carry the total box body to move, The control device is electrically connected with the sorting mechanism and can control the start and stop of the sorting mechanism according to the control signal. 3.The smart sorting bin based on image recognition sorting according to claim 2, characterized in that, The state monitoring device includes: Ultrasonic device, alarm lamp and smoke alarm; wherein, The ultrasonic device is arranged in the total box body and in each garbage storage box, can monitor the height of garbage in each garbage storage box through ultrasonic wave, and sends an electric signal when the height reaches a predetermined height, The alarm lamp and the smoke alarm are electrically connected with the ultrasonic device, and can send alarm light and smoke alarm prompt sound according to the electric signal of the ultrasonic device. 4.The smart sorting trash can based on image recognition sorting according to claim 1, wherein, The plurality of garbage storage boxes are at least two of the following: Household garbage storage box, kitchen waste garbage storage box, hazardous waste garbage storage box and other garbage storage box. 5.The smart sorting bin based on image recognition sorting according to claim 1, wherein, The visual identification classification device includes: Camera, convolutional neural network and control signal output module; wherein The camera is arranged at the garbage receiving mouth of the garbage receiving box side of the total box body, and can collect the garbage image of received garbage from the garbage receiving mouth; The convolutional neural network is communicatively connected with the camera, can take the garbage image collected by the camera as input, extract features and classify the garbage image, and determine the category of received garbage; The control signal output module is communicatively connected with the convolutional neural network, and can send the corresponding sorting control signal to the sorting mechanism according to the category of received garbage output by the convolutional neural network. 6.The smart sorting bin based on image recognition sorting according to claim 5, characterized in that, The convolutional neural network includes: First convolution layer, first pooling layer, second convolution layer, second pooling layer, full connection layer and regression layer; wherein, The first convolution layer can take the garbage image collected by the camera as input, perform first convolution processing to extract low-level features to obtain initial feature map, The first pooling layer, connected with the first pooling layer, can perform first down-sampling processing on the initial feature map output by the first convolutional layer to obtain a first compressed feature map; The second convolutional layer, connected with the first pooling layer, can perform second convolutional processing on the first compressed feature map output by the first pooling layer to extract high-level semantic features to obtain an intermediate feature map; The second pooling layer, connected with the second convolutional layer, can perform second down-sampling processing on the intermediate feature map output by the second convolutional layer to obtain an intermediate compressed feature map; The full connection layer, connected with the second pooling layer, is used for flattening processing on the second compressed feature map output by the second pooling layer, and weighting fusion and comprehensive discrimination on the high-dimensional semantic features contained therein, to obtain a classification discrimination feature vector or a classification score vector corresponding to each garbage category; The regression layer, connected with the full connection layer, is used for normalized regression processing on the classification discrimination feature vector or the classification score vector corresponding to each garbage category output by the full connection layer, to output a probability distribution of each category to which the garbage belongs, and finally determine the category to which the garbage belongs according to the category with the maximum probability value in the probability distribution.

7. The smart sorting bin based on image recognition sorting according to claim 6, characterized in that, In the convolutional neural network, The first convolutional layer adopts a 3×3 standard convolutional structure; The first pooling layer adopts a 2×2 maximum pooling structure; The second convolutional layer adopts a depth separable convolutional or a hollow convolutional structure; The second pooling layer adopts a 2×2 maximum pooling structure; The full connection layer adopts a full connection structure with ReLU activation; The regression layer adopts a Softmax regression structure.

8. The smart sorting bin based on image recognition sorting according to claim 6 or 7, characterized in that, The convolutional neural network further comprises: A saliency feature enhancement unit connected between the first convolutional layer and the first pooling layer, which can perform saliency feature enhancement on the garbage main body region in the intermediate feature map output by the first convolutional layer to obtain an initial feature map with enhanced garbage main body region features, and output the initial feature map to the first pooling layer; And / or, the convolutional neural network further comprises: An attention module connected before or inside the full connection layer, which can perform attention enhancement on the high-dimensional features of the second compressed feature map processed by the full connection layer to enhance the feature selection capability of the full connection layer. 9.The smart sorting bin based on image recognition sorting according to claim 8, characterized in that, The saliency feature enhancement unit comprises: A saliency feature extraction module, a saliency weight generation module and a saliency weighted fusion module; wherein, The saliency feature extraction module, connected with the first convolutional layer, can perform feature convergence on the initial feature map output by the first convolutional layer in the spatial dimension to obtain a saliency response map; The saliency weight generation module, connected with the saliency feature extraction module, can perform normalization and non-linear mapping processing on the saliency response map obtained by the saliency feature extraction module to generate a saliency weight map; The saliency weighted fusion module, connected with the saliency weight generation module and the first convolutional layer respectively, can expand the saliency weight map generated by the saliency weight generation module in the channel dimension, and multiply it element by element with the initial feature map output by the first convolutional layer to obtain an initial feature map with enhanced garbage main body region features. The attention module adopts an SE channel attention module or a CBAM attention module.

10. The smart sorting bin based on image recognition sorting according to any one of claims 1-7, characterized in that, The sorting mechanism comprises: The garbage falling partition plate, the servo drive device, the transmission device, the rotating tray mechanism and the control device; wherein, The garbage falling partition plate is movably arranged between the lower end outlet of the garbage receiving box and the plurality of garbage storage boxes below in the general box body; The driving shaft of the servo drive device is connected with the garbage falling partition plate through the transmission device, and can drive the garbage falling partition plate to open and close; The rotating tray mechanism is arranged at the bottom of the general box body, and a plurality of mounting positions are arranged in a ring shape on the rotating tray mechanism, and one garbage storage box is arranged on each mounting position, and each garbage storage box can be driven to rotate to below the garbage falling partition plate; The control device is electrically connected with the visual recognition and classification device, the rotating tray mechanism and the servo drive device respectively, can control the servo drive device according to the corresponding sorting control signal sent by the visual recognition and classification device, and control the rotating tray mechanism to drive the corresponding garbage storage box to rotate to below the garbage falling partition plate.

Images