Intelligent garbage can based on CNN and voice interaction and system thereof

Abstract

The application discloses an intelligent garbage can based on CNN and voice interaction and a system thereof, and belongs to the technical field of environmental protection equipment. In view of the technical defects of low classification accuracy, large mechanical vibration, poor storage efficiency, single interaction and lack of remote management of the existing intelligent garbage can, the application comprises an aluminum profile frame, a component layer, a CoreXY two-axis sliding table, a scissor mechanism, a mechanical claw, a holder, a rudder, four types of independent garbage can bodies, an ultrasonic sensor and a pouring opening cover plate; a matching system comprises an image acquisition module, a CNN image recognition module, a motion control module, a voice interaction module, a full-load monitoring module, a data transmission module and a terminal management platform. The CNN module is based on a YOLOv11 model and is trained semi-supervisedly, and the CoreXY sliding table cooperates with the scissor mechanism to realize stable grabbing; the ultrasonic sensor triggers a full-load alarm, voice interaction assists classification, and the Internet of Things supports remote operation and maintenance. The application realizes automatic and accurate garbage classification, efficient storage and intelligent management, and is suitable for the application scene of environmental protection of smart cities.

Description

Technical Field

[0001] This invention relates to the field of environmental protection equipment, and in particular to a smart trash can and system based on CNN and voice interaction. Background Technology

[0002] With the comprehensive implementation of my country's waste sorting policy and the in-depth implementation of the green development concept, intelligent waste sorting equipment has become a research hotspot in the environmental protection field. Existing intelligent trash cans mostly focus on basic functions such as automatic lid opening, sensor lighting, and simple sterilization, but their ability to automatically and accurately sort waste is seriously insufficient, making it difficult to achieve autonomous identification, grabbing, and classification of waste.

[0003] The following core shortcomings exist in related technologies both domestically and internationally: Low level of intelligent sorting: It relies heavily on manual sorting and disposal. The few devices with recognition functions use traditional image algorithms, which have low accuracy in recognizing similar waste or complex backgrounds and cannot meet the needs of recognizing multiple types of household waste. Mechanical structural defects: Traditional sorting machines use complex structures such as cross-shaped optical axes, which have many parts, are complicated to install and maintain, and are prone to lateral forces and vibrations during movement, resulting in unstable garbage grabbing, easy dropping, and poor positioning accuracy; Low storage efficiency: Ordinary trash cans do not have a trash compression function, and recyclable trash is fluffy and takes up a lot of space, requiring frequent emptying, resulting in low resource utilization; Lack of monitoring and management: There is a lack of accurate full-load monitoring mechanism, making it impossible to issue real-time warnings; there is no Internet of Things linkage capability, making it impossible to remotely monitor equipment status and collect and classify data, resulting in low efficiency in sanitation dispatching; Poor user experience: It only has basic waste disposal functions and lacks voice Q&A and guidance functions. When the public has weak knowledge of waste sorting, it is easy to make mistakes in disposal.

[0004] In summary, existing technologies cannot simultaneously meet the core requirements of high recognition accuracy, high mechanical stability, high storage efficiency, intelligent interaction, and remote management. Therefore, a smart trash can solution integrating deep learning, precision mechanics, and IoT technologies is needed to address the aforementioned technical pain points. Summary of the Invention

[0005] Purpose of the invention: To overcome the shortcomings of existing technologies such as low accuracy in waste sorting, large mechanical vibration, poor grasping stability, low storage efficiency, simple interaction, and lack of remote management capabilities, and to provide an intelligent trash can and system based on CNN and voice interaction, which realizes automatic waste identification, accurate grasping and classification, efficient compression, full load warning, voice interaction Q&A, and remote monitoring and management, thereby improving the accuracy and efficiency of waste sorting and adapting to smart city and environmental protection application scenarios.

[0006] Technical solution: To solve the above-mentioned technical problems, according to one aspect of the present invention, more specifically, a smart trash can based on CNN and voice interaction, including an aluminum profile frame, a component layer, a CoreXY two-axis slide, a scissor mechanism, a mechanical claw, a gimbal, a servo motor, a trash can body, an ultrasonic sensor, a disposal port, and a cover. The aluminum profile frame serves as the main support structure for the equipment, with overall dimensions of 400mm×400mm×600mm. It is secured using L-shaped and T-shaped aluminum profile bolts, ensuring a stable structure. The component layer is fixed to the upper part of the aluminum profile frame, employing a compact layout to integrate electronic components for image acquisition, control, and communication. The CoreXY two-axis slide is mounted below the component layer and consists of linear guides, guide sliders, stepper motors, synchronous pulleys, idler pulleys, and synchronous belts. The linear guides are fixed to the aluminum profile frame, and the guide sliders are slidably connected to the linear guides. Two stepper motors are fixed to the aluminum profile frame and drive the three guide sliders to move through two synchronous pulleys, ten idler pulleys, and two synchronous belts, following the displacement formula. The two motors work together to achieve high-precision, high-speed movement of the slider, with no lateral force and minimal vibration; The scissor mechanism connects the CoreXY two-axis slide table and the mechanical gripper. The joint nodes are reinforced with aluminum columns to reduce lifting vibration, improve gripping stability, and realize the vertical lifting of the mechanical gripper. The gimbal is located below the scissor mechanism and is driven by two servo motors: one drives the gimbal to rotate from 0° to 180°, and the other drives the gimbal to pitch from 0° to 90°, adapting to the dumping angle requirements of waste of different sizes and weights. The trash can body is located at the bottom of the aluminum profile frame and is divided into four independent collection bins: recyclable, hazardous, kitchen waste, and other. The recyclable trash can has a built-in high-thrust DC electric push rod compression mechanism, with an effective compression rate of 85%. Each trash can body is equipped with an ultrasonic sensor on the top, with a detection range of 20mm-3300mm and a detection depth of 50mm-340mm. When trash is continuously detected at a depth of 110mm, a full load alarm is triggered. The top of the aluminum profile frame is provided with an inlet and a cover plate. The inlet has a size of 110mm×110mm. The cover plate is hinged to the top of the aluminum profile frame and is located above the component layer. It can be opened and closed to facilitate component maintenance.

[0007] Meanwhile, the present invention provides an intelligent trash can system based on CNN and voice interaction, including an image acquisition module, a CNN image recognition module, a motion control module, a voice interaction module, a full load monitoring module, a data transmission module, and a terminal management platform; Image acquisition module: A high-resolution camera on the side of the component layer to capture images of the shape, color, and texture of the waste inside the disposal port in real time, and transmit them to the CNN image recognition module; CNN Image Recognition Module: Integrated into the component layer, it adds a segmentation dataset module based on the YOLOv11 basic model, adopts a semi-supervised learning algorithm, and is trained through the CurriculumLabeling strategy; the confidence of pseudo-labels is selected according to the gradient of 20%-100%, and the parameters are initialized in each iteration to avoid the accumulation of wrong labels. It outputs garbage category and coordinate information, accurately distinguishes four types of household waste, and the recognition accuracy reaches more than 95%. Motion control module: Electrically connected to CoreXY two-axis slide, scissor mechanism, servo motor, and DC electric actuator; receives instructions from CNN image recognition module, calculates motor rotation angle, drives mechanical claw to move directly above the garbage, controls scissor mechanism lifting and lowering, mechanical claw grabbing, and then completes garbage dumping by gimbal rotation and pitching, synchronously controlling the operation of recyclable waste compression mechanism; Voice interaction module: Integrated on the outside of the aluminum profile frame, with built-in voice recognition and broadcasting unit, linked to the touch screen, responding to garbage classification consultation commands, answering questions in real time and displaying text guidance; Full load monitoring module: consists of 4 ultrasonic sensors, which monitor the capacity of each trash can in real time and upload data after triggering an alarm; Data transmission module: IoT communication unit, which realizes two-way data transmission between trash cans and terminal management platform, uploads data on disposal volume, full load status, and recognition accuracy, and receives remote control and algorithm update instructions; Terminal management platform: Real-time monitoring of equipment operation, statistical classification of data, and remote updating of CNN algorithm models to support sanitation dispatching decisions.

[0008] Beneficial effects: This invention utilizes a high-resolution camera and advanced CNN algorithm, employing deep learning to analyze the multi-dimensional features of waste, such as shape, color, and texture, through a CNN model, thereby accurately distinguishing different types of waste.

[0009] This invention employs a high-thrust DC electric pusher to compress recyclable waste, combined with a well-designed waste bin size, achieving an effective compression rate of 85% and significantly improving the storage efficiency of recyclable waste. Simultaneously, the optimized waste sorting mechanism, such as the CoreXY structure two-axis slide, reduces mechanical parts, lowers vibration, and improves the positioning accuracy and speed of the grippers, enabling rapid grabbing and sorting of waste.

[0010] This invention employs a multi-degree-of-freedom gimbal design, using servo motors to control the gimbal's pitch and rotation, precisely aligning it with four types of trash cans. It supports multi-angle dumping in complex scenarios, adapting to trash of different sizes and weights. The trash cans are equipped with touchscreen displays and voice interaction modules, providing residents with intuitive operation guidance and real-time Q&A. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the structure of the present invention; Figure 2 This is a front cross-sectional view of the present invention; Figure 3 This is a schematic diagram of the CoreXY two-axis slide and mechanical gripper in this invention; Figure 4 This is a schematic diagram of the stepper motor and synchronous belt in this invention; Figure 5 For the present invention Figure 4 A schematic diagram of the structure viewed from below.

[0012] Explanation of reference numerals in the attached diagram: 1. Aluminum profile frame; 2. Component layer; 3. CoreXY two-axis slide; 4. Scissor mechanism; 5. Mechanical gripper; 6. Gimbal; 7. Servo motor; 8. Trash can body; 9. Ultrasonic sensor; 10. Disposal port; 11. Cover plate; 31. Linear guide rail; 32. Guide rail slider; 33. Stepper motor; 34. Synchronous pulley; 35. Idler pulley; 36. Synchronous belt. Detailed Implementation

[0013] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0014] Reference Figures 1-5 A smart trash can based on CNN and voice interaction, with the following specific implementation: Frame construction: 400mm×400mm×600mm aluminum profiles are used and fixed with L-shaped and T-shaped connectors and bolts to form a stable aluminum profile frame 1; a 110mm×110mm insertion port 10 is reserved at the top and a hinged cover plate 11 is provided to ensure smooth opening and closing; Component layer installation: The control board, communication module, image acquisition module (high resolution camera) and other electronic components are compactly fixed on the upper part of the aluminum profile frame 1 to form component layer 2, and the circuit is neatly arranged and protected. CoreXY two-axis slide assembly: Three linear guide rails 31 are fixed parallel to each other on the frame below component layer 2; two stepper motors 33 are symmetrically fixed on both sides of the frame, paired with two synchronous pulleys 34 and ten idler pulleys 35, wound with two synchronous belts 36, and connected to the three guide rail sliders 32; the coordinated motion of the motors is adjusted to ensure that the slider movement accuracy and speed meet the requirements. Displacement formula; Installation of the grabbing and dumping mechanism: The top of the scissor mechanism 4 is fixed to the guide rail slider 32, and the bottom is connected to the mechanical claw 5. The joint nodes are reinforced with aluminum columns. The gimbal 6 is installed below the scissor mechanism 4 and is equipped with two servo motors 7, which are adjusted to rotate (0°-180°) and tilt (0°-90°) angles respectively to ensure that the dumping covers the four types of garbage bins. Deployment of trash cans and sensors: The bottom of the aluminum profile frame 1 is fixed with four independent trash can bodies 8, and the recyclable bin has a built-in DC electric push rod compression mechanism; each trash can is equipped with an ultrasonic sensor 9 on the top, with calibrated detection range (20mm-3300mm) and trigger depth (110mm). Interactive module installation: The voice interaction module and touch screen are fixed on the outside of the aluminum profile frame 1, and the wiring is connected and the function is debugged.

[0015] A smart trash can system based on CNN and voice interaction, workflow implementation: Image acquisition and recognition: After the user disposes of garbage through the disposal port 10, a high-resolution camera acquires garbage images in real time and transmits them to the CNN image recognition module; the module extracts shape, color and texture features through the YOLOv11 model, and analyzes the data using an algorithm trained by a semi-supervised CurriculumLabeling strategy to output garbage category and coordinate information; Automatic grabbing and sorting: The motion control module receives the recognition result and drives the CoreXY two-axis slide 3 to move the mechanical claw 5 to directly above the garbage; the scissor mechanism 4 descends and the mechanical claw 5 closes to grab the garbage; the slide moves the garbage to the top of the corresponding garbage bin, and the gimbal 6 rotates and tilts via the servo motor 7 to accurately dump the garbage into the corresponding bin; after recyclable garbage is put in, the compression mechanism automatically starts to compress; Full load monitoring and early warning: Ultrasonic sensor 9 monitors the depth of garbage in the trash can in real time. When garbage is continuously detected at a depth of 110mm, a full load alarm is triggered and uploaded to the terminal management platform through the data transmission module. Voice interaction service: When users touch the screen or ask questions by voice, the voice interaction module recognizes the instructions, broadcasts the garbage classification rules and disposal instructions in real time, and displays the text on the screen simultaneously; Remote management and maintenance: The terminal management platform receives real-time equipment operation data, classification data, and full load status, and supports remote viewing, parameter adjustment, and CNN algorithm model updates. Sanitation departments can optimize collection routes based on the data.

[0016] Implementation of algorithm training for an intelligent trash can system based on CNN and voice interaction: Dataset construction: Collect a large number of images of household waste, covering different lighting, angles, and occlusion scenes, and label them with four categories: recyclable, hazardous, kitchen waste, and other. Model training: Based on the YOLOv11 basic model, a segmentation dataset module was added; semi-supervised learning was adopted, and the confidence of pseudo-labels was gradually increased from 20% to 100% through the CurriculumLabeling strategy. Parameters were initialized in each iteration, and training was carried out until the recognition accuracy stabilized at over 95%. Model porting: The trained CNN model is ported to the component layer control board to adapt to edge computing and achieve local real-time recognition.

[0017] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A smart trash can based on CNN and voice interaction, characterized in that, include: Aluminum profile frame (1), component layer (2), CoreXY two-axis slide (3), scissor mechanism (4), mechanical claw (5), gimbal (6), servo motor (7), trash can body (8), ultrasonic sensor (9), disposal port (10) and cover plate (11); The aluminum profile frame (1) is the main support structure of the equipment, and the component layer (2) is fixed on the upper part of the aluminum profile frame (1) to integrate control electronic components; The CoreXY two-axis slide (3) is installed below the component layer (2) and is used to drive the mechanical gripper (5) to move precisely; The scissor mechanism (4) connects the CoreXY two-axis slide (3) and the mechanical claw (5) to realize the lifting and lowering of the mechanical claw (5); The gimbal (6) is located below the scissor mechanism (4) and is driven by a servo motor (7). It can rotate and pitch to dump garbage. The trash can body (8) is located at the bottom of the aluminum profile frame (1) and is divided into four independent storage bins: recyclable, hazardous, kitchen waste, and other. Each of the trash can bodies (8) is equipped with an ultrasonic sensor on the top, with a detection range of 20mm-3300mm and a detection depth of 50mm-340mm. When trash is detected at a depth of 110mm, a full load alarm is triggered. The sensor is installed on the side of the component layer (2).

2. The intelligent trash can according to claim 1, characterized in that: The CoreXY two-axis slide (3) includes a linear guide (31), a guide slider (32), a stepper motor (33), a synchronous pulley (34), an idler pulley (35), and a synchronous belt (36). The linear guide rail (31) is fixed to the aluminum profile frame (1), and the guide rail slider (32) is slidably connected to the linear guide rail (31). The stepper motor (33) is fixed to the aluminum profile frame (1) and drives the guide rail slider (32) to move the mechanical claw (5) through the synchronous wheel (34), idler wheel (35) and synchronous belt (36).

3. The intelligent trash can according to claim 1, characterized in that: The joints of the scissor mechanism (4) are reinforced with aluminum columns to reduce shaking during lifting and lowering and improve the gripping stability of the mechanical claw (5).

4. The intelligent trash can according to claim 1, characterized in that: The number of servo motors (7) is 2. One drives the gimbal (6) to rotate left and right from 0° to 180°, and the other drives the gimbal (6) to pitch from 0° to 90°, adapting to the dumping needs of garbage of different sizes and weights.

5. The intelligent trash can according to claim 1, characterized in that: The trash can body (8) is equipped with a high-thrust DC electric push rod compression mechanism. The recyclable trash can is adapted to this compression mechanism, and the effective compression rate of the trash reaches 85%, improving storage efficiency.

6. The intelligent trash can according to claim 1, characterized in that: The top of the aluminum profile frame (1) is provided with a delivery port (10) and a cover plate (11), and the cover plate (11) is connected to the top of the hinged aluminum profile frame (1). The cover plate (11) is located above the component layer (2).

7. A system for an intelligent trash can based on CNN and voice interaction, using an intelligent trash can based on CNN and voice interaction as described in any one of claims 1-6, comprising an image acquisition module, a CNN image recognition module, a motion control module, a voice interaction module, a full load monitoring module, a data transmission module, and a terminal management platform; The image acquisition module is a high-resolution camera, fixed to the side of the component layer, which acquires images of the shape, color, and texture of the waste inside the disposal opening in real time, and transmits them to the CNN image recognition module. Its features include: The image acquisition module is a high-resolution camera on the side of the component layer (2), which collects images of the shape, color and texture of the garbage in the disposal port (10) in real time and transmits the image data to the CNN image recognition module. The CNN image recognition module is integrated into the component layer (2), adopts a semi-supervised learning algorithm and adds a segmentation dataset module based on the YOLOv11 basic model. It is trained by the CurriculumLabeling strategy, the pseudo-label confidence is selected according to the gradient of 20%-100%, the parameters are initialized in each iteration, and the garbage category and coordinate information are output to the motion control module. The motion control module is electrically connected to the CoreXY two-axis slide (3), scissor mechanism (4), servo motor (7) and DC electric push rod respectively. Its core is the CoreXY two-axis slide (3): adopting the CoreXY structure, it has fewer mechanical parts, is easy to install and maintain, and the parallel belt drive method has no lateral force and can reduce motion vibration. It consists of two stepper motors (33), two synchronous pulleys (34), ten idler pulleys (35), two synchronous belts (36), three linear guides (31), and three guide sliders (32). The stepper motors (33) are fixed to the aluminum profile frame (1). Through the synchronous pulleys (34), idler pulleys (35), and synchronous belts (36), the rotational displacement of the two motors and the displacement of the clamping platform are established to meet the requirements. The displacement formula demonstrates how the coordinated motion of two motors enables high-precision, high-speed movement of the slider. The motion control module receives instructions from the CNN image recognition module, calculates the motor rotation angle, drives the CoreXY two-axis slide (3) to move the mechanical claw (5) to directly above the garbage, controls the scissor mechanism (4) to lift and lower, the mechanical claw (5) to grab, and then drives the gimbal (6) to rotate and pitch through the servo motor (7) to complete the garbage dumping, while controlling the DC electric push rod compression mechanism inside the recyclable garbage bin body (8) to run. The voice interaction module is integrated on the outside of the aluminum profile frame (1), with a built-in voice recognition and broadcasting unit, responding to the user's garbage classification consultation instructions and answering questions in real time, and linking the touch screen to display text guidance synchronously; The full load monitoring module consists of four ultrasonic sensors (9) on the top of the trash can body (8), with a detection range of 20mm-3300mm and a detection depth of 50mm-340mm. When trash is detected continuously at a depth of 110mm, a full load alarm is triggered and uploaded to the terminal management platform. The data transmission module is an Internet of Things (IoT) communication unit that enables bidirectional data transmission between the smart trash can and the terminal management platform, uploading data on the amount of trash disposed of, full load status, and recognition accuracy, and receiving remote control commands. The terminal management platform is used to monitor equipment operation in real time, collect and classify data, and remotely update CNN algorithm models to support sanitation dispatching decisions.

8. The intelligent trash can according to claim 7, characterized in that: The CNN image recognition module is integrated into the component layer (2). The garbage images captured by the high-resolution camera are analyzed by the CNN algorithm to identify the shape, color and texture features of the garbage, thereby achieving accurate identification of the four types of garbage.

9. The intelligent trash can according to claim 7, characterized in that: The CNN image recognition module adopts a semi-supervised learning algorithm. Through the CurriculumLabeling strategy, the confidence of pseudo-labels is selected in a gradient of 20%-100%. Parameters are initialized in each iteration to avoid the accumulation of erroneous labels and adapt to complex backgrounds and similar garbage recognition scenarios.

10. The intelligent trash can according to claim 7, characterized in that: The voice interaction module is located on the outside of the aluminum profile frame (1). The voice interaction module answers questions in real time and assists users in completing the garbage sorting and disposal.

Images