Intelligent garbage classification device based on flexible mechanical arm claw and visual identification

Abstract

The invention discloses an intelligent garbage classification device based on a flexible mechanical arm claw and visual recognition. The intelligent garbage classification device comprises a box cover and a box body. Wherein a man-machine interaction module, a garbage identification module and a garbage sorting module are arranged on the box cover; the box body is a garbage accommodating module; the man-machine interaction module is used for sensing garbage; the garbage identification module is used for identifying the type of garbage; and the garbage sorting module is used for grabbing the garbage recognized by the garbage recognition module. Through a pneumatic working mode and a computer vision recognition system, accurate and intelligent garbage classification is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of part classification, and particularly relates to an intelligent garbage classification device based on a flexible robotic arm claw and visual recognition. Background Art

[0002] At present, garbage classification mainly relies on manual labor, which is time-consuming and labor-intensive, causing a significant social and economic burden. Therefore, there is a need for a fully automatic intelligent garbage classification product to solve this social contradiction. However, due to the diverse types, textures, shapes, and sizes of garbage, existing garbage classification products use traditional rigid robotic arm claws, which are relatively single in form, lack flexibility, and have low grasping stability. When discriminating and grasping small and soft garbage, large errors will occur, resulting in inaccurate and incomplete garbage classification.

[0003] Taking some mainstream intelligent trash can products on the market as an example, such as the Townew T1 series intelligent trash cans launched by Xiaomi's ecological chain enterprises, which mainly rely on infrared induction for automatic opening and built-in sealing technology, with a certain level of intelligence, but still limited to simple garbage collection operations. This product does not have visual recognition and automatic sorting capabilities, and users still need to classify the garbage themselves before putting it in, failing to truly solve common problems such as "misthrowing" and "mixed throwing". Taking some commercial sorting equipment with a rigid robotic arm structure as an example, such as the early model iTrash intelligent sorting bin used in some public places, this equipment uses traditional metal grippers, and the clamping strength cannot be adjusted according to the garbage material. When facing soft or small garbage (such as paper balls, fruit peels, etc.), problems such as failed clamping, crushed garbage, or inaccurate sorting often occur, affecting the user experience and sorting efficiency. In addition, although some intelligent garbage classification terminals put into use by the municipal government (such as the pilot equipment in a certain district of Hangzhou) are equipped with an image recognition module, due to their algorithm being based on static image matching, lacking deep learning capabilities and flexible processing structures, the recognition accuracy significantly decreases when facing complex scenarios (such as insufficient light, garbage deformation, or package entanglement, etc.).

[0004] Therefore, there is an urgent need to develop an automatic garbage classification device that integrates a flexible grasping structure and a visual intelligent recognition algorithm to adapt to domestic garbage of different shapes and textures, and improve the accuracy, flexibility, and system adaptability of intelligent sorting. Summary of the Invention

[0005] In order to overcome the above problems existing in the prior art, the purpose of the present invention is to provide an intelligent garbage classification device based on a flexible robotic arm claw and visual recognition, which can achieve accurate and intelligent garbage classification through a pneumatic working mode and a computer vision recognition system.

[0006] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0007] An intelligent garbage classification device based on a flexible robotic arm claw and visual recognition, comprising a box cover and a box body; wherein, a human-computer interaction module, a garbage recognition module, and a garbage sorting module are arranged on the box cover; the box body is a garbage accommodation module;

[0008] The human-computer interaction module is used to sense garbage;

[0009] The garbage recognition module is used to identify the type of garbage;

[0010] The garbage sorting module is used to grab the garbage identified by the garbage recognition module.

[0011] The human-computer interaction module includes an infrared sensor, a temperature sensor, a smoke sensor, and a ultrasonic sensor located inside the box cover, and a slidable open garbage inlet located in the middle of the box cover;

[0012] The garbage recognition module includes a 360-degree panoramic camera located below the slidable open garbage inlet, an LED light source perpendicular to the box cover, and a pattern recognition and matching algorithm module for identifying garbage.

[0013] The garbage sorting module includes a PLC control system, a flexible pneumatic robotic arm and a flexible pneumatic robotic claw located in front of the slidable open garbage inlet, and the flexible pneumatic robotic claw is installed at the end of the flexible pneumatic robotic arm.

[0014] The garbage accommodation module includes a garbage input accommodation bin, a recyclable accommodation bin, a kitchen waste accommodation bin, a hazardous waste accommodation bin, and other garbage accommodation bins located inside the box body;

[0015] Photoelectric induction devices are equipped at the mouths of each garbage bin (recyclable, kitchen waste, hazardous, other) to confirm whether the garbage has been successfully put in;

[0016] An electronic weighing module can be optionally equipped at the bottom of the garbage bin to record the weight of various types of garbage every day;

[0017] A slide-type drawable garbage bag bracket is arranged at the bottom of the bin body to facilitate subsequent cleaning and replacement.

[0018] When a human body (object) approaches above the box cover and the infrared sensor detects the human body (object), it sends an instruction to the slidable open garbage inlet to automatically open the inlet, and then puts the garbage into the inlet and falls into the garbage accommodation bin in the box body; the sensor network composed of the temperature sensor, the smoke sensor, and the ultrasonic sensor uploads the temperature, smoke concentration, and whether the bin is full of the garbage bin to the user terminal in real time through the 5G network to ensure the safety of users and facilitate garbage collection.

[0019] The infrared sensor, the temperature sensor, the smoke sensor, and the ultrasonic sensor jointly form a sensor network;

[0020] The above sensor communicates with the main control chip via CAN or RS485 bus, and uploads the real-time monitoring data to the cloud management platform through the embedded 5G module, facilitating subsequent scheduling and alarm processing.

[0021] The infrared sensor is installed in the middle of the upper edge of the box cover for identifying the approaching distance of the human body;

[0022] The ultrasonic sensor is embedded above the trash storage bin to judge whether it is full; the temperature sensor and the smoke sensor are arranged on the top of the box body for monitoring possible odor or fire risks.

[0023] The 360-degree panoramic camera is installed directly below the trash chute, adopting a spherical rotating pan-tilt or a ring panoramic structure, and can collect images from different angles;

[0024] The LED fill lights are arranged in a ring and embedded around the 360-degree panoramic camera, and are automatically turned on to ensure the image recognition quality in a dim environment;

[0025] After the 360-degree panoramic camera collects the image, it enters the edge computing module, calls the CNN-based recognition algorithm module, and outputs the recognition result (such as "recyclables", "kitchen waste", etc.); the image acquisition and recognition process includes image acquisition, grayscale processing, edge enhancement, CNN feature extraction, Softmax classification output, and classification instruction transmission;

[0026] After the trash chute is slid open to complete an opening and closing action, the LED light source is turned on, and at the same time, the 360-degree panoramic camera combines with the pattern recognition and matching algorithm module to identify the trash type and send a sorting instruction to the sorting module; if no trash is recognized in the trash storage bin, the LED light source and the 360-degree panoramic camera are turned off to facilitate energy saving.

[0027] The flexible pneumatic robotic arm is set in the front of the trash chute, adopting a bionic multi-joint structure, and the flexible pneumatic gripper is controlled by the pneumatic cavity driving module to rotate and extend;

[0028] The flexible pneumatic gripper adopts a three-finger silicone inflatable cavity design, and realizes different clamping forces by adjusting the air pressure;

[0029] The flexible pneumatic gripper realizes dynamic position correction and safe grasping through the built-in position sensor and negative pressure feedback;

[0030] The flexible pneumatic robotic arm receives the classification result output by the CNN-based recognition algorithm module through the PLC, executes the grasping and rotates above the target classification bin to complete the throwing action.

[0031] An operation method of an intelligent classification trash device based on a flexible robotic arm claw and visual recognition includes the following steps;

[0032] Data collection and preprocessing:

[0033] After the 360-degree panoramic camera captures the garbage images, the CNN-based recognition algorithm module performs preprocessing operations on the images, including resizing the image size (such as unifying it to 224×224); color standardization and histogram equalization; removing background noise (edge detection or Gaussian filtering can be selected);

[0034] Feature extraction:

[0035] Use the trained CNN network (such as MobileNet, ResNet or self-developed lightweight model) to extract high-level semantic features from the garbage images, and the processing is as follows:

[0036] Let the input image be I, and the feature vector extracted by the network be f = CNN(I);

[0037] Among them, CNN represents the feature extraction process after multi-layer convolution, pooling, batch normalization and ReLU non-linear processing;

[0038] Classification and matching:

[0039] Input the image feature vector f into the fully connected layer and the Softmax classifier to output the probability distribution of each garbage category: P(y|I) = Softmax(Wf + b);

[0040] Among them, W is the classification weight matrix and b is the bias term. The system sets a threshold θ. If the maximum probability is less than this value, it is determined as "uncertain garbage", which is manually processed by the user or sent to the mixing bin;

[0041] Post-processing and control logic:

[0042] The recognition result is transmitted to the PLC control module, and the corresponding flexible pneumatic robotic arm operation program is triggered for different garbage categories to complete the subsequent grasping and placing actions.

[0043] The initial training of the CNN-based recognition algorithm module uses the self-built garbage image dataset (capturing domestic garbage samples), and is augmented with some open-source data (such as TrashNet, WasteNet). At the same time, the system supports online updating of model parameters, and realizes incremental learning and adaptive optimization through continuous collection and manual annotation of samples on the terminal side.

[0044] The garbage sorting module includes a PLC control system, a flexible pneumatic robotic arm, and a flexible pneumatic robotic gripper located in front of the slidable garbage inlet. Among them, the flexible pneumatic robotic arm adopts a modular bionic multi-degree-of-freedom structure, which is composed of several flexible drive units connected in series. Each section includes a flexible outer shell made of high-flexibility elastic silicone or TPU material, with compressive and corrosion-resistant capabilities; an embedded pneumatic cavity with symmetrically arranged air chamber channels inside each section, which can generate bending or stretching deformation when inflated; an embedded angle sensor and a limiting elastic body to achieve motion feedback and attitude control; and a high-density fiber coating on the outside to enhance the overall tension strength and deformation recovery ability.

[0045] Each flexible arm segment is connected to a solenoid valve through a hose, and the control unit realizes the composite motion of bending, stretching, and rotation by adjusting the air pressure difference in each cavity.

[0046] The flexible pneumatic robotic arm is installed on the front inner cover part of the garbage inlet, about 50-100 mm in front of the 360-degree panoramic camera, and is located on the top fixed bracket.

[0047] The flexible pneumatic robotic arm is integrally integrated with the trachea and cable through a mechanical connection frame, which is convenient for disassembly and maintenance. Rotation and telescopic direction: The flexible pneumatic robotic arm can rotate within a range of 120° left and right in the horizontal plane and can telescopic grab within a range of 120 mm up and down in the vertical direction. Action obstacle avoidance logic: The position of the arm end is detected in real time through the arm end position sensor (magnetic encoder or gyroscope), and the obstacle avoidance control is completed in cooperation with visual feedback.

[0048] The flexible pneumatic robotic gripper is designed as a three-finger bionic clamping structure. Each finger is composed of a flexible inflatable cavity, and its shape is similar to a soft finger-like structure made of silicone material, which can naturally wrap and clamp according to the shape of the garbage. Anti-slip lining patterns or flexible films can be selected to enhance the grasping friction.

[0049] After receiving the sorting instruction sent by the garbage recognition module, the PLC control system controls the flexible pneumatic robotic arm and the flexible pneumatic robotic gripper to grab different types of garbage and accurately put them into the corresponding garbage storage bins. Recyclables are put into the recyclable storage bin, food waste is put into the food waste storage bin, hazardous waste is put into the hazardous waste storage bin, and other waste is put into the other waste storage bin.

[0050] Induction devices are respectively arranged at the inlets of the recyclable storage bin, food waste storage bin, hazardous waste storage bin, and other waste storage bin to detect whether the flexible pneumatic robotic arm and the flexible pneumatic robotic gripper have successfully put the garbage into the corresponding storage bin.

[0051] The induction device is a non-contact infrared sensor or a photoelectric reflection sensor, which is installed at the upper diversion port or the side wall position of each garbage storage bin. Whenever the flexible pneumatic robotic arm completes garbage clamping and moves above the target classification bin for dumping operation, if it detects that the garbage blocks or interrupts the induction signal through the input port, the system will determine that the classification dumping is successful for this time.

[0052] If no valid signal is detected and returned within the set time, the system will determine that the dumping fails, and can automatically trigger the prompt module, or control the flexible robotic arm to perform a grasping and dumping operation again; at the same time, this induction device can also be used as a secondary confirmation means for classification behavior, to verify the accuracy of the recognition result of the visual recognition module, and upload it to the upper system through the communication module for log recording and data analysis; the induction device can be linked with the voice broadcast module or the LED status indicator module to give a voice feedback prompt (such as "the garbage has been put into the recyclable bin") or light up the green status light after successful classification, so as to improve the user interaction experience and the visibility of system feedback.

[0053] Advantages of the present invention:

[0054] The present invention has a sensor network system composed of multiple sensors as nodes, which can sense the human body to automatically open and close the garbage bin cover and upload the temperature, smoke concentration, and whether the bin is full data in the garbage bin in real time, being more intelligent and safe, and facilitating management and cleaning.

[0055] The present invention only uses one 360-degree panoramic camera and works in a time-sharing manner, saving resources.

[0056] The present invention adopts a pattern recognition and matching algorithm module, which can accurately identify the types of garbage and improve the accuracy of subsequent garbage classification.

[0057] The present invention adopts a pneumatic flexible robotic arm claw, which can grasp and classify small and soft garbage, and improve the flexibility of grasping garbage. Description of the Drawings

[0058] Figure 1 It is a schematic diagram of the whole device of the present invention.

[0059] Figure 2 It is a schematic diagram of the back of the device of the present invention.

[0060] Figure 3 It is a schematic diagram of the front of the device of the present invention.

[0061] Figure 4 It is a schematic diagram of the flexible pneumatic robotic arm claw of the present invention.

[0062] Figure 5 It is a schematic diagram of the side of the device of the present invention. Detailed Embodiment

[0063] The present invention will be further described in detail below with reference to the accompanying drawings.

[0064] As Figures 1-5 shown, an intelligent garbage classification device based on a flexible robotic arm claw and visual recognition includes a box cover 1 and a box body 2. Among them, the box cover 1 includes a human-computer interaction module, a garbage recognition module, and a garbage sorting module; the box body 2 is a garbage storage module.

[0065] The human-computer interaction module is used to sense garbage;

[0066] The garbage recognition module is used to identify the type of garbage;

[0067] The garbage sorting module is used to grab the garbage after it is recognized by the garbage recognition module.

[0068] The human-computer interaction module includes an infrared sensor 3, a temperature sensor 4, a smoke sensor 5, and a ultrasonic sensor 6 located inside the box cover 1, which together form a sensor network, and a slidable and open garbage inlet 7 located in the middle of the box cover 1. When a human body (object) approaches above the box cover 1 and the infrared sensor 3 detects the human body (object), it sends an instruction to the slidable and open garbage inlet 7 to automatically open the inlet 7, and then garbage is put into the inlet 7 and falls into the garbage storage bin 12 in the box body 2; the sensor network composed of the temperature sensor 4, the smoke sensor 5, and the ultrasonic sensor 6 uploads the temperature, smoke concentration, and whether the bin is full of the garbage bin to the user terminal in real time through the 5G network to ensure the safety of users and facilitate garbage collection.

[0069] The garbage recognition module includes a 360-degree panoramic camera 8 located below the slidable and open garbage inlet 7, an LED light source 9 perpendicular to the box cover 1, and a pattern recognition and matching algorithm for identifying garbage. Only after the slidable and open garbage inlet 7 completes one opening and closing action, the LED light source is turned on, and at the same time, the 360-degree panoramic camera combines the pattern recognition and matching algorithm to identify the type of garbage and send a sorting instruction to the sorting module; if no garbage is recognized in the garbage storage bin 12, the LED light source and the 360-degree panoramic camera are turned off to save energy.

[0070] The garbage sorting module includes a PLC control system, a flexible pneumatic robotic arm 10, and a flexible pneumatic robotic claw 11 located in front of the slidable and open garbage inlet 7. When receiving the sorting instruction sent by the garbage recognition module, the PLC control system controls the flexible pneumatic robotic arm 10 and the flexible pneumatic robotic claw 11 to grab different types of garbage and accurately put them into the corresponding garbage storage bins. Recyclables are put into the recyclable storage bin 13, food waste is put into the food waste storage bin 14, hazardous waste is put into the hazardous waste storage bin 15, and other garbage is put into the other garbage storage bin 16.

[0071] The garbage containment module includes a waste disposal bin 12 located inside the box body 2 for containing the garbage put in through the garbage inlet 7, a recyclable bin 13 for containing recyclables put in by the flexible pneumatic robotic arm 10 and the flexible pneumatic robotic claw 11, a food waste bin 14 for containing food waste put in by the flexible pneumatic robotic arm 10 and the flexible pneumatic robotic claw 11, a hazardous waste bin 15 for containing hazardous waste put in by the flexible pneumatic robotic arm 10 and the flexible pneumatic robotic claw 11, and other waste bin 16 for containing other waste put in by the flexible pneumatic robotic arm 10 and the flexible pneumatic robotic claw 11.

[0072] The infrared sensor 3, temperature sensor 4, smoke sensor 5, and ultrasonic sensor 6 together form a sensor network;

[0073] The above sensors communicate with the main control chip via CAN or RS485 bus, and upload the real-time monitoring data to the cloud management platform through the embedded 5G module, facilitating subsequent scheduling and alarm processing.

[0074] The infrared sensor 3 is installed in the middle of the upper edge of the box cover 1 for identifying the approaching distance of the human body;

[0075] The ultrasonic sensor 6 is embedded above the waste disposal bin 12 to determine whether it is full; the temperature sensor 4 and the smoke sensor 5 are arranged on the top of the box body 2 for monitoring possible odors or fire risks.

[0076] The 360-degree panoramic camera 8 is installed directly below the garbage inlet 7, adopting a spherical rotating cloud platform or a circular panoramic structure, and can collect images from different angles;

[0077] The LED fill light 9 is arranged in a ring and is embedded around the 360-degree panoramic camera 8, and is automatically turned on to ensure the image recognition quality in a dim environment;

[0078] After the 360-degree panoramic camera 8 collects an image, it enters the edge computing module, calls the CNN-based recognition algorithm module, and outputs the recognition result (such as "recyclables", "food waste", etc.); the image collection and recognition process includes image collection, grayscale processing, edge enhancement, CNN feature extraction, Softmax classification output, and classification instruction transmission;

[0079] After the garbage inlet 7 that can be slid open completes an opening and closing action, the LED light source is turned on, and at the same time, the 360-degree panoramic camera combines with the pattern recognition and matching algorithm module to identify the type of garbage and send a sorting instruction to the sorting module; if no garbage is recognized in the waste disposal bin 12, the LED light source and the 360-degree panoramic camera are turned off, which is beneficial for energy conservation.

[0080] The flexible pneumatic robotic arm 10 is arranged in front of the garbage disposal opening 7. It adopts a bionic multi-joint structure and is controlled by a pneumatic chamber drive module to rotate and extend the flexible robotic claw.

[0081] The flexible robotic claw adopts a three-finger silica gel inflatable chamber design and adjusts the air pressure to achieve clamping with different forces.

[0082] The flexible robotic claw realizes dynamic position correction and safe grasping through built-in position sensors and negative pressure feedback.

[0083] The flexible pneumatic robotic arm 10 receives the classification result output by the CNN-based recognition algorithm module through the PLC, executes the grasping action, rotates above the target classification bin, and completes the disposal action. The maximum extended length of the flexible pneumatic robotic arm 10: 400 mm - 600 mm; the maximum grasping radius: arranged according to the position of the garbage bin, adapting to the space within 500 mm; the load-bearing capacity: the mass of the garbage clamped at one time ≤ 500 g; the reaction speed: the single-action time is 1 - 2 s, suitable for the volume of daily household or community garbage. The three fingers of the flexible pneumatic robotic claw 11 are symmetrically distributed, each finger is about 90 mm long, and is equipped with an independent air chamber; the clamping claw angle: the clamping range is adjustable between 0° and 90°; the adaptability to the grasping object: it can automatically adapt to irregular soft garbage such as oval, spherical, and sheet-shaped; the clamping force control: adjusts the air chamber pressure through a pneumatic controller to precisely control the clamping force and avoid damaging the garbage shape or grasping failure.

[0084] The working principle of the present invention:

[0085] When the user approaches the device, the infrared sensor 3 inside the box cover 1 detects the human approach signal, and the system sends an opening instruction to the slidable garbage disposal opening 7 by the control unit, automatically opening the garbage disposal opening. After the user throws the garbage, the garbage falls into the garbage disposal receiving bin 12 located in the box body 2.

[0086] At this time, the 360-degree panoramic camera 8 directly below the garbage disposal opening 7 starts to work under the illumination of the LED light source 9, collecting garbage image information. The image data is preprocessed by the local calculation module and input into the pattern recognition matching algorithm for recognition. This recognition algorithm is based on a convolutional neural network (CNN), extracts features from the garbage image, and outputs multi-classifications. The recognition results include category labels such as "recyclables", "kitchen waste", "hazardous waste", and "other waste".

[0087] The recognition result is transmitted to the PLC control system, and the control system schedules the flexible pneumatic robotic arm 10 and the flexible pneumatic robotic claw 11 to perform response operations according to the recognized garbage category. The flexible robotic arm 10 is pneumatically driven, adjusts the bending angle and spatial coordinates under the control instruction, and moves to the garbage position; the flexible robotic claw 11 adjusts the grasping force and the included angle according to the size and shape of the target object to complete the grasping action.

[0088] Subsequently, the robotic arm moves the garbage above the corresponding garbage storage bin and drops the garbage into the recyclable waste storage bin 13, the food waste storage bin 14, the hazardous waste storage bin 15, or the other waste storage bin 16. An induction device 17 is provided at each sorting bin opening to detect whether the garbage is successfully dropped. Once the induction device confirms that the dropping action is completed, the system marks this recognition and sorting behavior as "successful" and updates the local record log.

[0089] By setting the above induction device, the reliability and intelligence level of the garbage sorting link can be effectively improved, the closed-loop control of the whole process of garbage dropping can be ensured, and the system robustness can be enhanced.

[0090] In addition, the temperature sensor 4, the smoke sensor 5, and the ultrasonic sensor 6 in the device form a sensor network to monitor the internal environment status of the device in real time, such as abnormal temperature, smoke leakage, or full garbage bin, etc., and upload it to the cloud platform or the user terminal through the 5G communication module to achieve remote warning and intelligent scheduling.

[0091] To sum up, through the five steps of "perception - recognition - decision - execution - feedback", the present invention realizes the functions of automatic garbage recognition and precise classification, and has significant advantages such as high recognition accuracy, strong sorting flexibility, friendly user interaction, and stable system operation.

Claims

1. An intelligent garbage classification device based on a flexible robotic arm claw and visual recognition, characterized in that It includes a box cover (1) and a box body (2); among them, a human-computer interaction module, a garbage recognition module, and a garbage sorting module are arranged on the box cover (1); the box body (2) is a garbage accommodation module; The human-computer interaction module is used to sense the garbage; The garbage recognition module is used to identify the type of garbage; The garbage sorting module is used to grab the garbage after being recognized by the garbage recognition module.

2. The intelligent garbage classification device based on a flexible robotic arm claw and visual recognition according to claim 1, characterized in that, The human-computer interaction module includes an infrared sensor (3), a temperature sensor (4), a smoke sensor (5), a ultrasonic sensor (6) located inside the box cover (1), and a slidable and openable garbage inlet (7) located in the middle of the box cover (1); The garbage recognition module includes a 360-degree panoramic camera (8) located below the slidable and openable garbage inlet (7), an LED light source (9) perpendicular to the box cover (1), and a pattern recognition and matching algorithm module for identifying garbage; The garbage sorting module includes a PLC control system, a flexible pneumatic manipulator (10) and a flexible pneumatic gripper (11) located in front of the slidable and openable garbage inlet (7), and the flexible pneumatic gripper (11) is installed at the end of the flexible pneumatic manipulator (10).

3. The intelligent garbage sorting device based on a flexible robotic arm claw and visual recognition according to claim 2, wherein, The garbage accommodation module includes a garbage input accommodation bin (12), a recyclable accommodation bin (13), a kitchen waste accommodation bin (14), a hazardous waste accommodation bin (15), and other garbage accommodation bins (16) located inside the box body (2); Each garbage bin is equipped with a photoelectric induction device at the opening to confirm whether the garbage is successfully put in; An electronic weighing module can be optionally configured at the bottom of the garbage bin to record the weight of various types of garbage every day; A slide-type drawable garbage bag bracket is arranged at the bottom of the bin body to facilitate subsequent cleaning and replacement.

4. The intelligent garbage classification device based on a flexible robotic arm claw and visual recognition according to claim 3, characterized in that, The infrared sensor (3), the temperature sensor (4), the smoke sensor (5), and the ultrasonic sensor (6) jointly form a sensor network; The above sensors communicate with the main control chip through the CAN or RS485 bus, and upload the real-time monitoring data to the cloud management platform through the embedded 5G module, which is convenient for subsequent scheduling and alarm processing; The infrared sensor (3) is installed in the middle of the upper edge of the box cover (1) to identify the approaching distance of the human body; The ultrasonic sensor (6) is embedded above the garbage input accommodation bin (12) to judge whether it is full; the temperature sensor (4) and the smoke sensor (5) are arranged on the top of the box body (2) to monitor possible odor or fire risks; The 360-degree panoramic camera (8) is installed directly below the garbage inlet (7), and adopts a spherical rotating cloud platform or a circular panoramic structure, and can collect images from different angles; The LED fill light (9) is arranged in a circular pattern and is embedded around the 360-degree panoramic camera (8), and is automatically turned on to ensure the image recognition quality in a dim environment.

5. An intelligent garbage sorting device based on a flexible robotic arm claw and visual recognition according to claim 4, characterized in that After the 360-degree panoramic camera (8) collects the image, it enters the edge computing module, calls the CNN-based recognition algorithm module, and outputs the recognition result; After the slidable garbage chute (7) completes one opening and closing action, the LED light source is turned on. At the same time, the 360-degree panoramic camera (8) combines with the pattern recognition and matching algorithm module to identify the type of garbage and send a sorting instruction to the sorting module; if no garbage is recognized in the garbage bin (12), the LED light source and the 360-degree panoramic camera (8) are turned off.

6. The intelligent garbage classification device based on a flexible robotic arm claw and visual recognition according to claim 5, wherein The flexible pneumatic robotic arm (10) is arranged in front of the garbage chute (7) and adopts a bionic multi-joint structure, and the flexible pneumatic gripper (11) is controlled by the pneumatic chamber drive module to rotate and extend. The flexible pneumatic gripper (11) adopts a three-finger silicone inflatable chamber design and realizes clamping with different forces by adjusting the air pressure. The flexible pneumatic gripper (11) realizes dynamic position correction and safe grasping through the built-in position sensor and negative pressure feedback. The flexible pneumatic robotic arm (10) receives the classification result output by the CNN-based recognition algorithm module through the PLC, executes the grasping and rotates above the target classification bin to complete the dropping action.

7. An intelligent garbage classification device based on a flexible robotic arm claw and visual recognition according to claim 6, characterized in that, The flexible pneumatic robotic arm (10) adopts a modular bionic multi-degree-of-freedom structure. The whole is composed of several flexible drive units connected in series. Each section includes a flexible outer shell - made of high-flexibility elastic silicone or TPU material; an embedded pneumatic cavity - there are symmetrically arranged air cavity channels inside each section, and bending or stretching deformation can occur when inflated; an embedded angle sensor and a limit elastic body - realizing motion feedback and attitude control; and a high-density fiber coating layer is provided outside. Each flexible drive unit is connected to the solenoid valve through a hose, and the control unit realizes the composite motion of bending, stretching and rotating by adjusting the air pressure difference of each cavity. The flexible pneumatic robotic arm (10) is installed on the front inner cover part of the garbage chute (7), at a position 50-100 mm directly in front of the 360-degree panoramic camera (8), and is located on the top fixing bracket. The flexible pneumatic robotic arm (10) is integrally integrated with the trachea and cable through a mechanical connection frame. The flexible pneumatic robotic arm (10) can rotate within 120° left and right in the horizontal plane and can stretch and grasp within 120 mm up and down in the vertical direction. The flexible pneumatic gripper (11) is designed as a three-finger bionic clamping structure, and each finger is composed of a flexible inflatable cavity, and the shape is similar to a soft finger structure made of silicone material.

8. The operation method of an intelligent garbage classification device based on a flexible robotic arm claw and visual recognition according to any one of claims 1-7, characterized in that, It includes the following steps; Data acquisition and preprocessing: After the 360-degree panoramic camera (8) collects the garbage image, the CNN-based recognition algorithm module performs preprocessing operations on the image, including image size scaling; color standardization and histogram equalization; removing background noise. Feature extraction: Use the trained CNN network to extract high-level semantic features from the garbage image, and the processing is as follows: Let the input image be I, and the feature vector extracted by the network be f = CNN(I); Among them, CNN represents the feature extraction process after multi-layer convolution, pooling, batch normalization and ReLU non-linear processing; Classification and matching: Input the image feature vector f into the fully connected layer and the Softmax classifier, and output the probability distribution of each garbage category: P(y|I) = Softmax(Wf + b); Among them, W is the classification weight matrix, b is the bias term, and the system sets a threshold θ. If the maximum probability is less than this value, it is determined as "uncertain garbage", which is manually processed by the user or sent to the mixing bin; Post-processing and control logic: The recognition result is transmitted to the PLC control module, which triggers the corresponding operation procedures of the flexible pneumatic manipulator (10) for different types of garbage to complete subsequent grasping and placing actions.

9. The operating method of an intelligent garbage classification device based on a flexible robotic arm claw and visual recognition according to claim 8, characterized in that, The CNN-based recognition algorithm module is initially trained using a self-built garbage image dataset and augmented with some open-source data. At the same time, the system supports online updating of model parameters to achieve incremental learning and adaptive optimization through continuous acquisition and manual annotation of samples on the edge side; After receiving the sorting instruction issued by the garbage recognition module, the PLC control system controls the flexible pneumatic manipulator (10) and the flexible pneumatic gripper (11) to grasp different types of garbage and accurately place them into the corresponding garbage storage bins. Recyclables are placed in the recyclable storage bin (13), food waste is placed in the food waste storage bin (14), hazardous waste is placed in the hazardous waste storage bin (15), and other garbage is placed in the other garbage storage bin (16); Inductive devices are respectively arranged at the inlets of the recyclable storage bin (13), the food waste storage bin (14), the hazardous waste storage bin (15) and the other garbage storage bin (16) to detect whether the flexible pneumatic manipulator (10) and the flexible pneumatic gripper (11) have successfully placed the garbage into the corresponding storage bin.

10. The operating method of an intelligent garbage sorting device based on a flexible robotic arm claw and visual recognition according to claim 9, characterized in that, The inductive device is a non-contact infrared sensor or a photoelectric reflection sensor, which is installed at the upper diversion port or the side wall position of each garbage storage bin. Whenever the flexible pneumatic gripper (11) completes the garbage clamping and moves above the target classification bin for placing operation, if it detects that the garbage blocks or interrupts the inductive signal through the inlet, the system judges that the classification placement is successful; If no valid signal is detected within the set time, the system judges that the placement fails and can automatically trigger the prompt module or control the flexible manipulator (10) to perform a grasping and placing operation again; at the same time, this inductive device can also be used as a secondary confirmation means for classification behavior to verify the accuracy of the recognition result of the visual recognition module and upload it to the upper-level system through the communication module for log recording and data analysis; the inductive device can be linked with the voice broadcast module or the LED status indicator module to give a voice feedback prompt or light up the green status light after successful classification, improving the user interaction experience and the visibility of system feedback.

Images