Self-supervised mixed domain adaptive multi-garbage classification device and control and identification method

Abstract

The invention discloses a self-supervised mixed domain adaptive multi-garbage classification device and a control and identification method, the multi-garbage classification device comprises a frame main body, a multi-stage transmission unit, a compression unit, a rotary storage unit and an identification display unit, the multi-stage conveying unit, the compression unit and the rotary storage unit are sequentially arranged in the frame body from top to bottom, the multi-stage conveying unit is located on one side of the central axis of the frame body, and the compression unit is located on the other side of the central axis of the frame body. A feeding opening is formed in the top end of the frame body and located above the multi-stage conveying unit, balls are fixedly arranged at the four corners of the bottom end of the frame body, a middle supporting plate is arranged in the middle of the frame body, a lower bottom plate is arranged at the bottom of the frame body, and a control unit is arranged on one side of the frame body. According to the method, the garbage separation success rate and the compression efficiency are improved through the mixed domain adaptation technology, and the intelligent garbage classification requirements of complex scenes such as communities and schools can be met.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent environmental protection equipment, and in particular to a self-supervised hybrid domain adaptive multi-garbage classification device, control and identification method. Background Art

[0002] With the acceleration of urbanization and the improvement of environmental awareness, garbage classification has become an important measure to improve the living environment and promote resource recycling. In the field of garbage classification technology, traditional devices generally have bottlenecks such as poor adaptability to multiple scenarios, single detection dimension and lack of interactive feedback. Intelligent garbage classification devices, as the core equipment for automatic classification, are widely used in public places such as communities and schools. However, due to the wide variety and different forms of garbage, and the significant differences in garbage characteristics in different regions, higher requirements are placed on the recognition accuracy and environmental adaptability of classification devices, especially in crowded places, where garbage is frequently and randomly placed, which poses severe challenges to traditional devices.

[0003] Most existing equipment uses a single-channel transmission structure, which cannot cope with the stacking interference caused by the simultaneous input of multiple garbage, resulting in misjudgment of the recognition system due to occlusion or composite materials. Mainstream classification algorithms mostly rely on a single sensor modality (such as weight or optical detection), which is difficult to deal with the feature confusion caused by the diversity of garbage forms; for example, when transparent glass and liquid containers coexist, traditional single sensor detection solutions are difficult to accurately identify. Although some devices improve classification accuracy by introducing technical means such as spacing maintenance mechanisms or acoustic feature detection, they fail to solve the problem of multi-target parallel processing; at the same time, although existing visual recognition technology has improved classification accuracy to a certain extent, it has not effectively solved the problem of cross-domain feature migration.

[0004] The technical bottlenecks of traditional devices are mainly reflected in three aspects: first, the adaptability to multiple types of scenes is poor. The single-channel transmission structure cannot effectively cope with the situation where multiple garbage is put in at the same time, and it is easy to cause misidentification due to garbage stacking; second, the detection dimension is single. It is difficult to accurately identify composite material garbage relying on a single sensor modality. There is a lack of effective domain adaptation mechanism, resulting in insufficient generalization performance of the model in different scenarios; finally, there is a lack of interactive feedback. Most devices lack a real-time feedback system, and users cannot know the classification results and container status, resulting in blind spots in operation. These problems seriously restrict the actual application effect of garbage sorting devices in complex scenes. There is an urgent need for a composite solution that integrates multi-level physical separation, intelligent compression storage, and self-supervised hybrid domain adaptation visual detection, and breaks through the existing technical bottlenecks through the collaborative innovation of mechanical structure and algorithm.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In response to the problems in the related technology, the present invention proposes a self-supervised hybrid domain adaptive multi-garbage sorting device, control and identification method, which has the advantages of mechanical separation optimization, intelligent compression, dynamic identification feedback and modular architecture, and has significant advantages in classification efficiency, space utilization, accuracy and environmental adaptability, thereby solving the problems of misjudgment of garbage accumulation, cumbersome operation and high maintenance cost in the prior art.

[0007] To this end, the specific technical solution adopted by the present invention is as follows:

[0008] According to one aspect of the present invention, a self-supervised hybrid domain adaptive multi-garbage sorting device is provided, which comprises a frame body, a multi-stage transmission unit, a compression unit, a rotating storage unit and an identification display unit, wherein the multi-stage transmission unit, the compression unit and the rotating storage unit are sequentially arranged inside the frame body from top to bottom, and the multi-stage transmission unit is located on one side of the central axis of the frame body, and the compression unit is located on the other side of the central axis of the frame body; a delivery port is provided at the top of the frame body above the multi-stage transmission unit, ball bearings are fixedly provided at the four corners of the bottom end of the frame body, a middle support plate is provided in the middle of the frame body, a lower base plate is provided at the bottom of the frame body, and a control unit is provided on one side of the frame body.

[0009] Further, the multi-stage conveying unit includes a V-shaped conveyor belt assembly arranged on one side of the top of the frame body;

[0010] The V-shaped conveyor belt assembly includes two groups of symmetrically arranged composite profiles, and the two groups of composite profiles are in a V-shaped obtuse angle structure. The bottom ends of the two groups of composite profiles are provided with a plurality of linearly distributed V-shaped conveyor belt connectors, and one side of the bottom of the two groups of composite profiles is commonly connected to a vibration motor;

[0011] Both ends of the composite profile are provided with V-shaped conveyor belt rollers, and one end of two groups of V-shaped conveyor belt rollers located on one side of the vibration motor is provided with a first stepper motor.

[0012] Furthermore, the multi-stage conveying unit further comprises a horizontal conveyor belt assembly arranged at a side of the middle of the frame body away from the V-shaped conveyor belt assembly;

[0013] The horizontal conveyor belt assembly comprises two groups of symmetrically arranged horizontal conveyor belt rollers, the outer sides of the horizontal conveyor belt rollers are sleeved with horizontal conveyor belts, both ends of the horizontal conveyor belt rollers are provided with bearing seats, and a connecting profile is provided between the two groups of bearing seats;

[0014] A second stepper motor is arranged at the bottom end of the connecting profile, and a synchronous wheel is sleeved on the output end of the second stepper motor. The synchronous wheel is connected to a horizontal conveyor belt roller wheel away from the compression unit side through a synchronous belt.

[0015] Further, the compression unit includes an electric push rod assembly arranged on one side of the middle part of the frame body;

[0016] The electric push rod assembly includes a rear support plate arranged on one side of the middle part of the frame body, an electric push rod is arranged on one side of the rear support plate, an active compression plate is arranged at the output end of the electric push rod, a first optical axis is penetrated at both ends of the active compression plate, an optical axis support frame is arranged at one end of the first optical axis, and a linear motion bearing fixedly connected to the active compression plate is arranged in the middle of the first optical axis;

[0017] A front support plate is commonly provided at the other ends of the two groups of first optical axes, and optical axis support seats matching the first optical axes are provided on both sides of the front support plate, and a passive compression plate is provided in the middle of the front support plate.

[0018] Further, the compression unit also includes a sliding door assembly arranged in the middle of the frame body and below the electric push rod assembly;

[0019] The sliding door assembly includes two sets of sliding door tracks symmetrically arranged at the top of the middle support plate, one side of the sliding door track is provided with a sliding groove, and the inner side of the sliding groove is provided with a sliding door;

[0020] A sliding door upper rack is arranged on one side of the top of the sliding door, and a third stepping motor is arranged on one side of the sliding door track away from the multi-stage transmission unit. The output end of the third stepping motor is arranged with a sliding door gear matched with the sliding door upper rack.

[0021] Further, the rotating storage unit includes a rotating chassis arranged at the bottom of the frame body, a plurality of trash can bases distributed in a circumference are opened at the top of the rotating chassis, a trash can body is arranged on the top of the trash can base, and a driven gear is arranged at the bottom of the rotating chassis; a fourth stepper motor is arranged at one side of the rotating chassis below the compression unit, and a driving gear meshing with the driven gear is arranged at the bottom of the output end of the fourth stepper motor;

[0022] An optical axis fastener is arranged through the middle of the rotating chassis, a second optical axis is arranged through the middle of the optical axis fastener, a middle fixing piece matched with the optical axis fastener is sleeved on the outer side of the middle of the second optical axis, tapered roller bearings are sleeved on the top and bottom ends of the second optical axis, and a fixing piece is arranged on the outer side of the tapered roller bearing.

[0023] According to another aspect of the present invention, a self-supervised hybrid domain adaptive multi-waste classification control method is also provided, the control method comprising:

[0024] SI, separate the input garbage through the V-type conveyor belt assembly at a differential speed, and dynamically adjust the differential ratio through the control unit;

[0025] SII, pause the horizontal conveyor assembly, use the camera to collect a static image and transmit it to the control unit;

[0026] SIII, the control unit completes the garbage classification identification and outputs the result to the control board when the confidence reaches the threshold;

[0027] SIV, select the operation mode according to the classification result. When it is identified as recyclable garbage, execute the recyclable garbage mode, and execute the operations of closing the sliding door, compressing the electric push rod, rotating the rotating chassis and opening the sliding door in sequence; when it is identified as non-recyclable garbage, execute the non-recyclable garbage mode, and execute the operations of rotating the rotating chassis and opening the sliding door in sequence;

[0028] SV, updates the garbage classification information and capacity status through the display screen, and triggers a shutdown warning when the capacity of the garbage bin exceeds the threshold.

[0029] According to another aspect of the present invention, a self-supervised hybrid domain adaptive multi-garbage classification identification method is also provided, the identification method comprising:

[0030] S1, obtain multi-view images of garbage through the camera and pre-process them;

[0031] S2. Use the hybrid domain adaptation garbage recognition model to extract image features, realize spatial selective scanning and channel switching, and use the transfer learning and incremental learning framework to update the model parameters;

[0032] S3, based on the hybrid domain adaptive garbage identification model, output the garbage classification results and confidence, and synchronously update them to the display screen and control unit;

[0033] Among them, S2 includes:

[0034] S21. Build a hybrid domain adaptation garbage recognition model to improve feature extraction capabilities by embedding channel attention modules and contrastive learning loss functions;

[0035] S22, processing feature maps based on spatial selective scanning mechanism and channel exchange strategy, and realizing spatial selective scanning and channel exchange by calculating cross-domain feature entropy;

[0036] S23. Combine the transfer learning pre-trained weights and incremental learning framework to dynamically update the model parameters and improve the recognition performance of garbage categories.

[0037] Furthermore, a hybrid domain adaptation garbage recognition model is constructed, and the feature extraction capability is improved by embedding the channel attention module and contrastive learning loss function, including:

[0038] S211, extracting shallow features of source domain and target domain images through two consecutive convolution blocks, and generating preliminary feature maps through maximum pooling downsampling;

[0039] S212, using depthwise separable convolution to process the preliminary feature map, measuring the spatial consistency of the source domain feature and the target domain feature map by cosine similarity, and generating a similarity matrix;

[0040] S213, equally dividing the source domain feature map and the target domain feature map along the channel dimension, calculating the channel-level similarity after exchanging some channels, and generating features that are spatially aligned with the channels;

[0041] The expression of channel-level similarity is:

[0042] ;

[0043] In the formula, S channel represents channel-level similarity; [ , ] represents the feature vector of the concatenation of segments 1 and 2 of the source domain feature map and segments 3 and 4 of the target domain feature map; [ , ] represents the feature vector obtained by concatenating segments 1 and 2 of the target domain feature map and segments 3 and 4 of the source domain feature map; ||·|| represents the norm of the vector.

[0044] Furthermore, based on the spatial selective scanning mechanism and the channel exchange strategy, the feature map is processed, and the spatial selective scanning and channel exchange are realized by calculating the cross-domain feature entropy, including:

[0045] S221, using self-attention to independently enhance the intra-domain features, generating source dominant features and target dominant features through cross-attention, and achieving soft alignment of features;

[0046] S222, passing the source domain features, the source dominant features, the target dominant features, and the target domain features to the next stage, and repeating the feature extraction and alignment process;

[0047] S223, introduce entropy loss in shallow, middle and deep layers, calculate cross-domain feature entropy through edge activation function, and strengthen cross-domain consistency of multi-level features;

[0048] The expression of cross-domain feature entropy is:

[0049] ;

[0050] ;

[0051] In the formula, H t→ts (l) represents the entropy value of the dominant characteristic of the source in the first stage; k represents the proportionality coefficient; σ m represents the edge activation function; Z t→ts (l,k) Indicates the dominant characteristic of the source in stage l; Z t (1,k)represents the target domain features of the first stage; L entropy represents the total entropy loss; N represents the number of feature layers; l represents the feature layer index; H t→st (l) Represents the entropy value of the target dominant feature in the lth stage.

[0052] The beneficial effects of the present invention are:

[0053] (1) Multi-stage separation and parallel processing capabilities improve efficiency: The present invention adopts a two-stage collaborative design of differential separation of V-type conveyor belt components and detection of horizontal conveyor belt components, and adopts V-type conveyor belts and adaptive differential control to achieve efficient physical separation of multiple garbage. At the same time, the V-type conveyor belt detects the garbage stacking density in real time through infrared sensors, and combines with dual stepper motor differential drive to automatically optimize the differential ratio according to volume differences or material friction coefficient differences. In addition, silicone corrugated belts are used in both the horizontal conveyor belt component and the V-type conveyor belt component, and a vibration motor is added to the bottom of the carbon fiber composite aluminum profile. When the separation force is insufficient, high-frequency vibration assistance is triggered to ensure that the garbage enters the detection area one by one, which significantly improves the separation efficiency and stacking separation success rate, while reducing energy consumption and extending the life of the conveyor belt.

[0054] (2) Compression function optimizes storage space utilization: A linkage system of an electric push rod assembly and a sliding door assembly is designed for recyclable garbage. The active compression plate and the passive compression plate work together to compress the loose garbage into shape. Compared with traditional devices without compression function, the present invention can significantly reduce the volume of garbage, extend the full load cycle of the garbage bin, and reduce the frequency of users dumping garbage. It is especially suitable for high load requirements in public scenes.

[0055] (3) Intelligent identification and dynamic interaction enhance classification accuracy: The present invention realizes the coordination of algorithm identification and mechanical control based on a dual control system. The wide-angle camera provides high-precision image acquisition and combines with the autonomous classification algorithm to determine the type of garbage in real time. The rotating storage unit dynamically matches the corresponding garbage bin, and the display screen synchronously feeds back the classification results and full load status. Compared with existing equipment that relies on manual selection or a single sensor, the misjudgment rate is greatly reduced and the transparency of user operation is improved.

[0056] (4) Modular structural design improves environmental adaptability: The present invention adopts a composite aluminum profile frame and an independent PCB control unit, integrating multiple motors, sensors and power supply systems. It has a compact structure and strong anti-interference ability. The design of the V-shaped conveyor belt and the rotating base of the trash can reduces friction loss. Combined with a one-button start-stop switch and a wide-angle camera, it is suitable for a variety of complex indoor and outdoor environments. Compared with traditional fixed or loosely assembled equipment, it has both scalability and long-term operation stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 is a structural schematic diagram of a multi-waste classification device for self-supervised hybrid domain adaptation according to an embodiment of the present invention;

[0059] Figure 2 is a rear view of a multi-waste classification device for self-supervised hybrid domain adaptation according to an embodiment of the present invention;

[0060] Figure 3 is a right view of a multi-waste classification device for self-supervised hybrid domain adaptation according to an embodiment of the present invention;

[0061] Figure 4 is a structural schematic diagram of a V-shaped conveyor belt assembly in a multi-waste sorting device for self-supervised hybrid domain adaptation according to an embodiment of the present invention;

[0062] Figure 5 is a structural schematic diagram of a horizontal conveyor belt assembly in a multi-waste sorting device for self-supervised hybrid domain adaptation according to an embodiment of the present invention;

[0063] Figure 6 is a structural schematic diagram of an electric push rod assembly in a multi-waste sorting device for self-supervised hybrid domain adaptation according to an embodiment of the present invention;

[0064] Figure 7 is a structural schematic diagram of a middle sliding door assembly of a multi-waste classification device for self-supervised hybrid domain adaptation according to an embodiment of the present invention;

[0065] Figure 8 is a structural schematic diagram of a rotating storage unit in a multi-waste classification device for self-supervised hybrid domain adaptation according to an embodiment of the present invention;

[0066] Fig. 9 is a schematic diagram of a local structure of a rotating storage unit in a multi-waste classification device for self-supervised hybrid domain adaptation according to an embodiment of the present invention;

[0067] Fig.10 is a flowchart of a self-supervised hybrid domain adaptive multi-waste classification control method according to an embodiment of the present invention;

[0068] Fig.11 is a specific implementation diagram of a multi-garbage classification control method for self-supervised hybrid domain adaptation according to an embodiment of the present invention;

[0069] Fig.12 It is a flowchart of a self-supervised hybrid domain adaptive multi-garbage classification identification method according to an embodiment of the present invention.

[0070] In the figure:

[0071] 1. Frame body; 101. Inlet; 102. Ball bearing; 103. Middle support plate; 104. Lower bottom plate; 105. Infrared ranging sensor; 2. Multi-stage transmission unit; 201. First stepper motor; 202. Composite profile; 203. V-type conveyor belt roller; 204. V-type conveyor belt connector; 205. Vibration motor; 206. Synchronous belt; 207. Horizontal conveyor belt roller; 208. Horizontal conveyor belt; 209. Second stepper motor; 210. Bearing seat; 211. Connecting profile; 212. Synchronous wheel; 3. Compression unit; 301. Electric push rod; 302. Rear support plate; 303. Active compression plate; 304. Optical axis support frame; 305. Linear motion bearing; 306. First stepper motor; 207. Horizontal conveyor belt roller; 208. Horizontal conveyor belt; 209. Second stepper motor; 210. Bearing seat; 211. Connecting profile; 212. Synchronous wheel; 3. Compression unit; 301. Electric push rod; 302. Rear support plate; 303. Active compression plate; 304. Optical axis support frame; 305. Linear motion bearing; 306. An optical axis; 307, an optical axis support seat; 308, a passive compression plate; 309, a front support plate; 310, a third stepper motor; 311, a sliding door gear; 312, a sliding door upper rack; 313, a sliding door track; 314, a slide groove; 315, a sliding door; 4, a rotating storage unit; 401, a fourth stepper motor; 402, a driving gear; 403, a rotating chassis; 404, a trash can body; 405, a trash can base; 406, a driven gear; 407, a second optical axis; 408, a middle fixing piece; 409, an optical axis fastener; 410, a fixing piece; 411, a tapered roller bearing; 5, an identification display unit; 501, a display screen; 502, a camera; 6, a control unit; 7, a photoelectric emergency stop sensor. DETAILED DESCRIPTION

[0072] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in the field should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0073] According to an embodiment of the present invention, a self-supervised hybrid domain adaptive multi-garbage classification device, control and identification method is provided.

[0074] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1-Figure 9As shown, according to one embodiment of the present invention, a self-supervised hybrid domain adaptive multi-waste classification device is provided, the self-supervised hybrid domain adaptive multi-waste classification device comprises a frame body 1, a multi-stage transmission unit 2, a compression unit 3, a rotation storage unit 4 and an identification display unit 5, the multi-stage transmission unit 2, the compression unit 3 and the rotation storage unit 4 are sequentially arranged inside the frame body 1 from top to bottom, and the multi-stage transmission unit 2 is located on one side of the central axis of the frame body 1, and the compression unit 3 is located on the other side of the central axis of the frame body 1;

[0075] A delivery port 101 is provided at the top of the frame body 1 above the multi-stage conveying unit 2, balls 102 are fixedly provided at the four corners of the bottom end of the frame body 1, a middle support plate 103 is provided in the middle of the frame body 1, a lower base plate 104 is provided at the bottom of the frame body 1, and a control unit 6 is provided on one side of the frame body 1.

[0076] Specifically, Figure 1-Figure 3 As shown, a self-supervised hybrid domain adaptive multi-garbage sorting device proposed in the present invention includes a frame body 1, a multi-stage transmission unit 2, a compression unit 3, a rotating storage unit 4 and an identification display unit 5; the multi-stage transmission unit 2, the compression unit 3 and the rotating storage unit 4 are sequentially installed in the frame body 1 from top to bottom; a delivery port 101 is arranged on the frame body 1 above the V-shaped conveyor belt assembly, and a ball 102 is fixed at the bottom of the frame body 1 to facilitate the movement of the device; the identification display unit 5 includes a display screen 501 and a camera 502, a display screen 501 is installed above the frame body 1, and a camera 502 required for detection and identification is installed on the frame body 1 located above the horizontal conveyor belt assembly; the control unit 6 includes a control board (the model of the control board in this embodiment is STM32F446RCT6) and a development board (the model of the development board in this embodiment is Jetson nano); the rotating storage unit 4 is equipped with four garbage can bodies 404, which are respectively used to store recyclable garbage, hazardous garbage, kitchen waste and other garbage.

[0077] It should be noted that the first stepper motor 201, the vibration motor 205, the second stepper motor 209, the electric push rod 301, the third stepper motor 310, the fourth stepper motor 401, the display screen 501, the camera 502 are electrically connected to the control unit 6. This is the prior art and will not be elaborated here.

[0078] After power is turned on, the display screen 501 plays the garbage classification promotion video in a loop. After a plurality of garbage is put into the delivery port 101, the garbage passes through the V-shaped conveyor belt assembly.

[0079] Specifically, in the above embodiment, the multi-stage conveying unit 2 includes a horizontal conveyor belt assembly and a V-shaped conveyor belt assembly. The V-shaped conveyor belt assembly is composed of two conveyor belts arranged at an angle of 110° to 130°, and is differentially driven by two sets of first stepper motors 201 (in this embodiment, the first stepper motor is a 42 series two-phase stepper motor, model 42HS402004D, with a rated torque of 0.4N·m and a peak current of 1.5A) (in this embodiment, the differential ratio of the two sets of first stepper motors is set to 1:1.2~1.8), and the surface of the conveyor belt is coated with a polyurethane anti-slip layer ( In this embodiment, the Shore hardness of the polyurethane anti-slip layer is 80A, and the friction coefficient is μ=0.6), and a vibration motor 205 is integrated at the bottom (the model of the vibration motor in this embodiment is YZD206, the frequency is 30Hz±5%, and the amplitude is 1.5mm); the horizontal conveyor belt assembly is driven by a second stepper motor 209 (the second stepper motor in this embodiment is a 42 series two-phase stepper motor, model 42HS402004D), and a one-way baffle is provided at the end (in this embodiment, the inclination angle of the one-way baffle to the vertical direction is set to 15°), and the transmission speed is 0.25~1.25m / s.

[0080] Specifically, in the above embodiment, the compression unit 3 includes an electric push rod assembly and a sliding door assembly, the electric push rod assembly: the electric push rod 301 (the specification of the electric push rod in this embodiment is 24V / 1000N) drives the active compression plate 303, and a film pressure sensor is provided at the front end of the compression plate (the range of the film pressure sensor in this embodiment is 0~1500N, and the accuracy is ±1%); the sliding door assembly: the third stepper motor 310 (the third stepper motor in this embodiment is a 42 series two-phase stepper motor, model 42HS402004D) controls the linear movement of the sliding door through a gear rack mechanism (the modulus of the gear rack mechanism in this embodiment is 1.5) (the stroke of the sliding door in this embodiment is 120mm);

[0081] Specifically, in the above embodiment, the rotating storage unit 4 includes a fourth stepper motor 401 (the fourth stepper motor in this embodiment is a 57 series three-phase stepper motor) driving a rotating chassis 403, which is positioned by a tapered roller bearing 411 and a second optical axis 407 (the diameter of the second optical axis in this embodiment is 17 mm), and 32 Hall positioning grooves are evenly distributed in the circumference of the chassis;

[0082] Specifically, in the above embodiment, the identification display unit 5 includes a camera 502 (in this embodiment, the camera is a 4K wide-angle camera with a field of view of 120°), an infrared distance sensor 105 (in this embodiment, the specification of the infrared distance sensor is a detection accuracy of ±2mm);

[0083] Control unit 6: A control board (the model of the control board in this embodiment is STM32F446RCT6) and a development board (the model of the development board in this embodiment is Jetson nano) are used to form a dual-core architecture, and the control board is connected to the actuator via a CAN bus (the baud rate of the CAN bus in this embodiment is 1Mbps);

[0084] Frame body 1: In this embodiment, a European standard 2020 aluminum profile body is used, and a ball 102 (the diameter of the ball in this embodiment is 25 mm) and a 6S lithium battery compartment (the capacity of the 6S lithium battery compartment in this embodiment is 10000 mAh) are provided at the bottom.

[0085] In one embodiment, the multi-stage conveying unit 2 includes a V-shaped conveyor belt assembly disposed on one side of the top of the frame body 1;

[0086] The V-shaped conveyor belt assembly includes two sets of symmetrically arranged composite profiles 202, and the two sets of composite profiles 202 are in a V-shaped obtuse structure. The bottom ends of the two sets of composite profiles 202 are provided with a plurality of linearly distributed V-shaped conveyor belt connectors 204, and one side of the bottom of the two sets of composite profiles 202 is commonly connected to a vibration motor 205;

[0087] V-shaped conveyor belt rollers 203 are disposed at both ends of the composite profile 202 , and a first stepper motor 201 is disposed at one end of two groups of V-shaped conveyor belt rollers 203 located on one side of the vibration motor 205 .

[0088] In one embodiment, the multi-stage conveying unit 2 further includes a horizontal conveyor belt assembly disposed in the middle of the frame body 1 on one side away from the V-shaped conveyor belt assembly;

[0089] The horizontal conveyor belt assembly includes two groups of symmetrically arranged horizontal conveyor belt rollers 207, the outer sides of the horizontal conveyor belt rollers 207 are sleeved with horizontal conveyor belts 208, both ends of the horizontal conveyor belt rollers 207 are provided with bearing seats 210, and a connecting profile 211 is provided between the two groups of bearing seats 210;

[0090] A second stepper motor 209 is disposed at the bottom end of the connecting profile 211 , and a synchronous wheel 212 is sleeved on the output end of the second stepper motor 209 . The synchronous wheel 212 is connected to a horizontal conveyor belt roller 207 on a side away from the compression unit 3 via a synchronous belt 206 .

[0091] Specifically, Figure 3 and Figure 4As shown, the V-type conveyor belt assembly includes two first stepper motors 201, the output shafts of the two first stepper motors 201 are connected to two groups of V-type conveyor belt rollers 203 through couplings, so as to drive the two groups of V-type conveyor belt rollers 203 to rotate (in this embodiment, the reduction ratio between the output shaft of the first stepper motor 201 and the V-type conveyor belt rollers 203 is set to 1:2), and the conveyor belts on both sides are fixed and formed at an angle of 110°~130° through the composite profile 202 (in this embodiment, the composite profile adopts CF-AL6061 carbon fiber composite aluminum profile) to drive the rotation, and the surface of the composite profile 202 is provided with a pressure sensor and a silicone corrugated belt as a conveyor belt; the first stepper motor 201 adopts In the closed-loop control mode (the subdivision accuracy of the closed-loop control mode in this embodiment is 1600 pulses / revolution), the control panel adjusts the differential ratio in real time (the differential ratio of the two first stepper motors in this embodiment is 1:1.2~1:5), and the corresponding conveyor belt linear speed range is 0.25m / s and 0.3~1.25m / s, so that the stacked garbage generates a 5~15N lateral separation force due to the speed difference. The surface of the silicone corrugated belt is provided with a fishbone-shaped anti-skid pattern, and the bottom of the composite profile 202 is integrated with a vibration motor 205 to assist in separating the adhered garbage; the V-shaped conveyor belt connector 204 is forged with 7075-T6 aluminum alloy and fixed to the European standard 2020 aluminum profile of the frame body 1 by M5 high-strength bolts. After testing, the component can separate 5 pieces of stacked garbage (the size of the garbage in this embodiment is 50~300mm and the weight is 10~500g) within 2 seconds, with a separation success rate of 98.5% and a continuous operating life of ≥50,000 times.

[0092] Specifically, the separated garbage falls onto the horizontal conveyor belt assembly in sequence. Figure 5 As shown, the horizontal conveyor belt assembly is driven by the second stepper motor 209 to rotate the synchronous belt 206 through the rotating synchronous wheel 212. The rotation of the synchronous belt 206 drives the horizontal conveyor belt roller 207 to rotate through engagement. The horizontal conveyor belt roller 207 drives the horizontal conveyor belt 208 to rotate through friction. The horizontal conveyor belt 208 also adopts a silicone corrugated belt. The surface of the silicone corrugated belt is provided with a fishbone-shaped anti-slip pattern, so that the horizontal conveyor belt 208 can achieve transportation through the friction generated by the surface and the garbage.

[0093] Specifically, before transporting garbage, the horizontal conveyor assembly remains stationary. When garbage falls onto the horizontal conveyor assembly, the camera 502 detects it, and the V-shaped conveyor assembly remains stationary to prevent more garbage from entering the horizontal conveyor assembly. When the camera 502 completes the detection, the horizontal conveyor assembly starts to move, and the garbage is transported into the compression unit 3.

[0094] Specifically, the camera 502 acquires images on the horizontal conveyor belt assembly in real time and feeds them back to the development board. The development board processes, identifies and analyzes the acquired photos and sends the analysis results to the control board, which controls the movement of the V-shaped conveyor belt assembly, the horizontal conveyor belt assembly, the subsequent compression unit 3 and the rotary storage unit 4.

[0095] Specifically, when the development board analyzes that the garbage is recyclable (such as paper cups, iron cups, and plastic bottles), the control board controls the compression unit 3 to move. Figure 6 and Figure 7 In the electric push rod assembly, the electric push rod 301 is fixed to the rear support plate 302 by bolts, and a passive compression plate 308 is fixed in front of the electric push rod 301 to increase the compression surface; the active compression plate 303 is fixed on both sides by linear motion bearings 305 (the linear motion bearings in this embodiment are linear bearing trimming flange motion bearings LUU). In order to enhance stability, the rear end of the first optical axis 306 (the diameter of the first optical axis in this embodiment is 8 mm) is fixed by the optical axis support frame 304, and the front end is fixed to the front support plate 309 by the linear optical axis support seat 307 (the model of the linear optical axis support seat in this embodiment is SHFSHF8), so that the active compression plate 303 can move along the first optical axis 306; the passive compression plate 308 is fixed to the front support plate 309 by bolts.

[0096] In one embodiment, the compression unit 3 includes an electric push rod assembly disposed on one side of the middle portion of the frame body 1;

[0097] The electric push rod assembly includes a rear support plate 302 arranged on one side of the middle part of the frame body 1, an electric push rod 301 is arranged on one side of the rear support plate 302, an active compression plate 303 is arranged at the output end of the electric push rod 301, a first optical axis 306 is penetrated at both ends of the active compression plate 303, an optical axis support frame 304 is arranged at one end of the first optical axis 306, and a linear motion bearing 305 fixedly connected to the active compression plate 303 is arranged in the middle of the first optical axis 306;

[0098] A front support plate 309 is commonly provided at the other end of the two groups of first optical axes 306 , and optical axis support seats 307 matching the first optical axes 306 are provided on both sides of the front support plate 309 , and a passive compression plate 308 is provided in the middle of the front support plate 309 .

[0099] In one embodiment, the compression unit 3 further includes a sliding door assembly disposed in the middle of the frame body 1 and below the electric push rod assembly;

[0100] The sliding door assembly includes two sets of sliding door rails 313 symmetrically arranged at the top of the middle support plate 103, a sliding groove 314 is opened on one side of the sliding door rail 313, and a sliding door 315 is arranged inside the sliding groove 314;

[0101] A sliding door upper rack 312 is arranged on the top side of the sliding door 315, and a third stepper motor 310 is arranged on the side of the sliding door track 313 away from the multi-stage transmission unit 2. The output end of the third stepper motor 310 is provided with a sliding door gear 311 matched with the sliding door upper rack 312.

[0102] Specifically, Figure 7 As shown, the sliding door assembly is driven by the third stepping motor 310 to rotate the sliding door gear 311; the sliding door gear 311 is meshed with the upper rack 312 of the sliding door, thereby driving the sliding door 315 to move forward and backward in the sliding door track 313. The electric push rod assembly is fixed to the frame body 1 through the rear support plate 302 and the front support plate 309; the sliding door assembly is fixed to the middle support plate 103 of the frame body 1 through the sliding door track 313.

[0103] Specifically, Figure 6 As shown, after the compression unit 3 receives the information from the control board, the third stepper motor 310 in the electric push rod assembly starts working, driving the sliding door 315 to move toward one end of the horizontal conveyor belt assembly. At this time, the sliding door 315 covers the drop opening on the middle support plate 103, thereby closing the downward sliding channel for the garbage; the electric push rod 301 starts to move, driving the active compression plate 303 to move forward to compress the recyclable garbage; when the compression is completed, the electric push rod 301 retreats, and the third stepper motor 310 rotates in the opposite direction, driving the rack 312 on the sliding door to retreat, and driving the sliding door 315 to move away from one end of the horizontal conveyor belt assembly. At this time, the sliding door 315 moves away, and the recyclable garbage passes through the drop opening on the middle support plate 103 under the action of gravity and automatically falls into the rotating storage unit 4.

[0104] Specifically, in the above embodiment, the electric push rod assembly is provided with a pressure feedback system: a thin film pressure sensor is arranged at the front end of the active compression plate 303 (the model of the thin film pressure sensor in this embodiment is FlexiForceA201), and the sampling rate is 1kHz; two-level compression thresholds are set: the first threshold of 500N triggers slow compression (the speed is reduced to 50%), and the second threshold of 800N triggers emergency stop protection; after the compression is completed, three vibration rebounds are performed (the frequency of the vibration rebound in this embodiment is 20Hz, and the amplitude is 1mm).

[0105] In one embodiment, the rotating storage unit 4 includes a rotating chassis 403 arranged at the bottom of the frame body 1, a plurality of trash can bases 405 distributed in a circumference are opened at the top of the rotating chassis 403, a trash can body 404 is arranged on the top of the trash can base 405, and a driven gear 406 is arranged at the bottom of the rotating chassis 403; a fourth stepping motor 401 is arranged on one side of the rotating chassis 403 below the compression unit 3, and a driving gear 402 meshing with the driven gear 406 is arranged at the bottom of the output end of the fourth stepping motor 401;

[0106] An optical axis fastener 409 is provided through the middle of the rotating chassis 403, a second optical axis 407 is provided through the middle of the optical axis fastener 409, a middle fixing piece 408 matching with the optical axis fastener 409 is sleeved on the outer side of the middle of the second optical axis 407, tapered roller bearings 411 are sleeved on the top and bottom ends of the second optical axis 407, and a fixing piece 410 is provided on the outer side of the tapered roller bearing 411.

[0107] Specifically, before the recyclable garbage falls, the control panel controls the rotating storage unit 4 to move according to the garbage type information received. Figure 8 and Fig. 9 The fourth stepping motor 401 in the rotating storage unit 4 drives the driving gear 402 to rotate, and the driving gear 402 is meshed with the driven gear 406. The top of the driven gear 406 is fixed with a rotating chassis 403; the entire rotating chassis 403 is fixed by a middle fixing member 408 and a bottom fixing member 410, and is tightly fixed to the second optical axis 407 through an optical axis fastener 409. The rotating chassis 403 rotates and drives the second optical axis 407 to rotate at the same time; the upper part of the second optical axis 407 is connected to the tapered roller bearing 408 at the top. 11 and the fixing piece 410 are fixed on the middle support plate 103 of the frame body 1, and the lower part of the second optical axis 407 is fixed on the lower bottom plate 104 of the frame body 1 through the fixing piece 410 at the bottom and the tapered roller bearing 411 at the bottom; the trash can body 404 is placed above the rotating chassis 403 for easy placement; a number of circumferentially distributed infrared ranging sensors 105 are installed on the middle support plate 103 of the frame body 1, which can detect the full load condition of the trash can body 404 to facilitate subsequent normal operation.

[0108] Specifically, after the rotating storage unit 4 rotates to a predetermined position, the garbage naturally falls into the corresponding garbage can body 404, and the display screen 501 displays the corresponding information of the garbage, including: garbage type, garbage quantity, whether the garbage has been put in, etc. When the infrared ranging sensor 105 detects that the garbage capacity in the garbage can body 404 exceeds three quarters of the garbage can capacity, the infrared ranging sensor 105 sends information to the control board, the control board controls all devices to suspend all work, and transmits the full load information to the development board, and the development board sends the full load information to the display screen 501 for display.

[0109] Specifically, in the above embodiment, the positioning system of the rotating storage unit 4 includes: a driver for the fourth stepper motor 401 (the driver model in this embodiment is a TMC5160 stepper motor driver), which is used to achieve 256 subdivision control; an electromagnetic locking mechanism (the holding torque of the electromagnetic locking mechanism in this embodiment is ≥2N·m), which is used to start after rotating into place; an RFID tag set on the trash can base 405 (the frequency is set to 13.56MHz in this embodiment), which is used to identify the type of trash can.

[0110] Specifically, the control unit 6 in the present invention is provided with three levels of safety protection: a) mechanical protection: a photoelectric emergency stop sensor 7 is provided at the entrance of the conveyor belt (the response time of the photoelectric emergency stop sensor in this embodiment is <50ms); b) electrical protection: leakage protection (the action current of the leakage protection in this embodiment is 30mA) and overload protection (the threshold current of the overload protection in this embodiment is 5A); c) data protection: AES256 is used to encrypt and transmit data, and local storage is retained for ≥30 days.

[0111] Specifically, the frame body 1 in the present invention adopts a modular design: the V-shaped conveyor belt assembly is installed through a quick-release interface (the quick-release interface in this embodiment uses M8 bolts), supporting 60°-150° angle adjustment; the trash can base is provided with an expansion interface (the expansion interface in this embodiment is Type C), supporting 26 types of garbage container adaptation; the top is integrated with a solar power supply interface (the solar power supply interface in this embodiment is an MPPT controller with a conversion efficiency ≥95%).

[0112] Specifically, in the present invention, the appearance photo data of the garbage is obtained by identifying the display unit 5, and the type of garbage is identified by the internal algorithm model of the development board (i.e., the self-supervised hybrid domain adaptive multi-garbage classification identification method in another embodiment); through the cooperation of the development board and the control board, the movement of the entire device is controlled to achieve the identification, classification and storage of multiple garbage. Compared with classification only through appearance data, volume data and weight data, the recognition accuracy is higher, and the precise cooperation between the upper computer and the lower computer of the device makes the classification more accurate. The present invention solves the industry pain points of low recognition accuracy and limited classification efficiency of traditional garbage classification equipment in complex scenarios through the deep integration of hardware collaboration and intelligent algorithms.

[0113] like Figure 10-11 According to another embodiment of the present invention, a self-supervised hybrid domain adaptive multi-garbage classification control method is also provided, and the control method includes:

[0114] SI, separate the input garbage through the V-type conveyor belt assembly, and dynamically adjust the differential ratio through the control unit;

[0115] SII, pause the horizontal conveyor assembly, use the camera to collect a static image and transmit it to the control unit;

[0116] SIII, the control unit completes the garbage classification identification and outputs the result to the control board when the confidence reaches the threshold;

[0117] SIV, select the operation mode according to the classification result. When it is identified as recyclable garbage, execute the recyclable garbage mode, and execute the operations of closing the sliding door, compressing the electric push rod, rotating the rotating chassis and opening the sliding door in sequence; when it is identified as non-recyclable garbage, execute the non-recyclable garbage mode, and execute the operations of rotating the rotating chassis and opening the sliding door in sequence;

[0118] SV, updates the garbage classification information and capacity status through the display screen, and triggers a shutdown warning when the capacity of the garbage bin exceeds the threshold.

[0119] Specifically, ① the V-shaped conveyor belt assembly is used to differentially separate multiple garbage that are thrown in at the same time, and the differential ratio is dynamically adjusted to 1:1.2~1:1.8 (automatically matched according to the material friction coefficient and volume difference);

[0120] ② When the horizontal conveyor assembly pauses for 3 seconds, the camera 502 collects a static image and transmits it to the development board (delay ≤ 50ms);

[0121] ③The development board completes the classification within 500ms and outputs the result to the control board when the confidence level is ≥0.85;

[0122] ④Select the operation mode according to the classification results:

[0123] Recyclable garbage mode: close the sliding door 315 (in this embodiment, the sliding door response time is set to <100ms) → start the 24V / 1000N electric push rod 301 to compress (in this embodiment, the electric push rod compression force is 200±5N, the stroke is 150mm) → rotate the rotating chassis 403 to the corresponding garbage bin (in this embodiment, the positioning accuracy is set to ±0.5°) → open the sliding door 315;

[0124] Non-recyclable garbage mode: directly rotate the rotating chassis 403 to the corresponding garbage bin (in this embodiment, the rotation angular velocity is set to 90° / s) → open the sliding door 315;

[0125] ⑤ The display screen 501 updates the classification information and the capacity status of the trash can in real time. When the infrared ranging sensor 105 detects that the capacity is ≥ 75%, a shutdown warning is triggered (the buzzing frequency is set to 2kHz in this embodiment).

[0126] like Fig.12 According to another embodiment of the present invention, a self-supervised hybrid domain adaptive multi-garbage classification identification method is also provided, and the identification method includes:

[0127] S1, obtaining multi-view images of garbage through camera 502 and pre-processing;

[0128] S2. Use the hybrid domain adaptation garbage recognition model to extract image features, realize spatial selective scanning and channel switching, and use the transfer learning and incremental learning framework to update the model parameters;

[0129] S3, based on the hybrid domain adaptive garbage identification model, output the garbage classification results and confidence, and synchronously update them to the display screen and control unit;

[0130] Specifically, ① obtaining multi-view images of the garbage through the camera 502, with a resolution of not less than 3840×2160 and a frame rate of ≥30fps;

[0131] ② Preprocessing the image, including deblurring (Wiener filter is used in this embodiment), illumination equalization (CLAHE algorithm is used in this embodiment) and background segmentation (MaskRCNN model is used in this embodiment);

[0132] ③ A hybrid domain adaptation garbage identification model (an improved YOLOv8 model is used in this embodiment) is used for feature extraction, the model is embedded with an SE attention module (the channel compression ratio of the SE attention module in this embodiment is 16:1) and a contrastive learning term is added to the loss function (the temperature coefficient of the contrastive learning loss in this embodiment is τ=0.07);

[0133] ④ Load the pre-trained weights through transfer learning (the ImageNet dataset is used in this embodiment), and use the incremental learning framework to update the fully connected layer parameters (the learning rate of the incremental learning optimizer is set to 1e-4 and the momentum coefficient is set to 0.9 in this embodiment);

[0134] ⑤ Output the classification results and confidence (the threshold of the model prediction confidence is set to 0.85 in this embodiment), and update them to the display screen 501 and the control panel simultaneously.

[0135] In one embodiment, the hybrid domain adaptation garbage recognition model is used to extract image features, realize spatial selective scanning and channel switching, and the transfer learning and incremental learning framework is used to update the model parameters, including:

[0136] S21. Build a hybrid domain adaptation garbage recognition model to improve feature extraction capabilities by embedding channel attention modules and contrastive learning loss functions;

[0137] S22, processing feature maps based on spatial selective scanning mechanism and channel exchange strategy, and realizing spatial selective scanning and channel exchange by calculating cross-domain feature entropy;

[0138] S23. Combine the transfer learning pre-trained weights and incremental learning framework to dynamically update the model parameters and improve the recognition performance of garbage categories.

[0139] Specifically, a) a spatial selective scanning mechanism is constructed to extract cross-domain spatial features through depthwise separable convolution (in this embodiment, the convolution kernel size is set to 3×3 and the stride is set to 1);

[0140] b) Implementing the channel exchange strategy, dividing the feature map into four segments along the channel dimension and then calculating the channel-level similarity (in this embodiment, the weight update is triggered when the cosine similarity is greater than 0.7);

[0141] c) An entropy loss layer is introduced before the detection head to calculate the cross-domain feature entropy value, which is expressed as:

[0142] ;

[0143] Where, L entropy represents the entropy loss value; i=2,4,6 represents the selected feature layer index; E[] represents the mathematical expectation; σ m represents the edge activation function (in this embodiment, the MarginReLU function is used, and the threshold θ=0.2); F (i) Represents the feature map output of different layers;

[0144] d) When updating the model parameters every week, 10% of the historical data is retained (FIFO strategy is adopted in this embodiment), and the accuracy of new category recognition is improved by ≥15%.

[0145] In one embodiment, a hybrid domain adaptation garbage recognition model is constructed to improve feature extraction capability by embedding a channel attention module and a contrastive learning loss function, including:

[0146] S211, extracting shallow features of source domain and target domain images through two consecutive convolution blocks, and generating preliminary feature maps through maximum pooling downsampling;

[0147] S212, using depthwise separable convolution to process the preliminary feature map, measuring the spatial consistency of the source domain feature and the target domain feature map by cosine similarity, and generating a similarity matrix;

[0148] S213, equally dividing the source domain feature map and the target domain feature map along the channel dimension, calculating the channel-level similarity after exchanging some channels, and generating features that are spatially aligned with the channels;

[0149] The expression of channel-level similarity is:

[0150] ;

[0151] In the formula, S channel represents channel-level similarity; [ , ] represents the feature vector of the concatenation of segments 1 and 2 of the source domain feature map and segments 3 and 4 of the target domain feature map; [ , ] represents the feature vector obtained by concatenating segments 1 and 2 of the target domain feature map and segments 3 and 4 of the source domain feature map; ||·|| represents the norm of the vector.

[0152] Specifically, the present invention adopts a hybrid domain adaptation method combined with Mamba-Transformer to achieve real-time monitoring of garbage in the open world, so that the model can adapt to the regionality of garbage samples. s ={(x s ,B s ,C s )} represents a set of labeled images in the source domain, where B s and C s Represent the source image x s The corresponding bounding box and category label of . In the target domain, there are D t ={x t}, including N t An image x excluding boundaries and class labels t The goal of identification is to use D s and D t Information development domain-adaptive open-world spam detection algorithms.

[0153] Specifically, Step 1: shallow feature extraction stage, the model receives images from the source domain and the target domain as input. First, low-level features are extracted through two convolution blocks. Each convolution block contains a convolution layer, batch normalization (BatchNorm) and a ReLU activation function. The size of the input image is (H, W, C), where H and W represent the height and width of the image, respectively, and C represents the number of channels. The convolution kernel size is 3×3, and the step size is 1. The size of the output feature map is (H, W, C1), where C1 is the number of output channels of the first stage convolution layer. Use the maximum pooling layer for downsampling, the pooling kernel size is 2×2, and the step size is 2. The size of the output feature map is (H / 2, W / 2, C1), and its operation can be expressed as:

[0154] ;

[0155] In the formula, Represents the output of the first stage of the source domain and the target domain; ConvBlock represents the convolution block; x s,t Representing images of source and target domains, Represents the set of real numbers.

[0156] Specifically, in Step 2, the convolutional layer is used again to extract low-level spatial features, and the size of the output feature map is (H / 2, W / 2, C2), where C2 is the number of output channels of the second-stage convolutional layer. The maximum pooling layer is used for downsampling, and the size of the output feature map is (H / 4, W / 4, C2). The operation can be expressed as:

[0157] ;

[0158] In the formula, Represents the output of the second stage of the source domain and the target domain; the convolution block captures low-level spatial features such as edges and corners through the local receptive field, providing a basis for subsequent domain adaptation. Here, the inductive bias of the convolution is used to retain the local structure, avoiding the premature loss of fine-grained information in the deep network and laying the foundation for domain adaptation.

[0159] Specifically, Step 3: Advanced feature extraction and alignment. After the first two convolution stages, the feature maps of the source domain and the target domain are processed. , input the hybrid domain adaptation module composed of Mamba-Transformer. The hybrid domain adaptation module goes through the domain adaptation Mamba module, attention mechanism, and feature transfer and output in sequence. Specifically, the steps are as follows:

[0160] Specifically, Step 31: Domain Adaptive Mamba Module, the domain adaptive Mamba module contains spatial selective scanning and channel exchange dimension mixing to adaptively model domain changes. Specifically: The system is applied independently to the source and target feature maps through deep convolution, preserving detailed spatial cues.

[0161] Specifically, in Step 311, the spatial selective scanner calculates cross-domain spatial similarity: and target domain features The spatial consistency of the feature maps of the two domains is measured by cosine similarity through deep separable convolution to generate a similarity matrix. This process focuses on the shared spatial pattern to capture the local structural information that is crucial for domain adaptation. The expression of the similarity matrix is:

[0162] ;

[0163] Where DWConv represents the separable convolution operation, Indicates the calculation of cosine similarity of feature vectors after one or two stages;

[0164] The feature maps are re-weighted based on similarity to enhance the shared spatial patterns in spatially consistent areas and suppress domain-specific noise. The weighted expression is:

[0165] ;

[0166] ;

[0167] Where ⊙ represents element-by-element multiplication; After weighting the representation similarity, the source domain output features in the third stage; Represents the output features of the source domain in the second stage; Represents the output of the target domain in the third stage after similarity weighting; Represents the output features of the target domain in the second stage;

[0168] The weighted features and focus on local structures shared across domains (such as texture and shape) enhance the feature fusion capability of the intermediate layer, thereby improving domain adaptation performance. Output aligned spatial features.

[0169] Specifically, Step 312: Channel exchange space model, input source domain feature map and target domain feature map It is divided into four sections along the channel dimension. , and calculate the channel-level similarity after exchanging some channels:

[0170] ;

[0171] Channel similarity vector S channel It is used to adjust channel weights and suppress domain-specific noise. The expression is:

[0172] ;

[0173] ;

[0174] Where, ChannelAttn represents the channel attention function; It indicates that the third stage adjusts the weights based on the channel similarity vector; Represents the output of the feature source domain and target domain after using cross attention.

[0175] Specifically, Step 313: Features of spatial and channel alignment and .

[0176] In one embodiment, processing feature maps based on a spatial selective scanning mechanism and a channel exchange strategy, and implementing spatial selective scanning and channel exchange by calculating cross-domain feature entropy includes:

[0177] S221, using self-attention to independently enhance the intra-domain features, generating source dominant features and target dominant features through cross-attention, and achieving soft alignment of features;

[0178] S222, passing the source domain features, the source dominant features, the target dominant features, and the target domain features to the next stage, and repeating the feature extraction and alignment process;

[0179] S223, introduce entropy loss in shallow, middle and deep layers, calculate cross-domain feature entropy through edge activation function, and strengthen cross-domain consistency of multi-level features;

[0180] The expression of cross-domain feature entropy is:

[0181] ;

[0182] ;

[0183] In the formula, H t→ts (l) represents the entropy value of the dominant characteristic of the source in the first stage; k represents the proportionality coefficient; σ m represents the edge activation function, which is used to suppress negative responses; Z t→ts (l,k) Indicates the dominant characteristic of the source in stage l; Z t (1,k) represents the target domain features of the first stage; L entropy represents the total entropy loss; N represents the number of feature layers; l represents the feature layer index, including shallow layer (l=2), middle layer (l=4) and deep layer (l=6); H t→st (l) Represents the entropy value of the target dominant feature in the lth stage.

[0184] Specifically, in Step 32, the features obtained through the spatial scanning mechanism are fed into the self-attention module, where the intra-domain features are independently enhanced to enhance spatial consistency:

[0185] ;

[0186] ;

[0187] The source domain key values ​​are , the target domain key is , used for internal feature refinement. SelfAtten represents the self-attention operation. The expression of the self-attention operation is as follows:

[0188] ;

[0189] Cross-attention generates source dominant feature Z t→s With the target dominant feature Z s→t Through cross-domain interaction of key-value pairs, soft-aligned feature distribution, and blurred domain boundaries:

[0190] ;

[0191] Here Crossattn represents the cross attention operation, and the cross attention operation is as follows:

[0192] ;

[0193] Where CrossAttn represents the result of the cross attention operation; Q represents the query matrix; K represents the key matrix; V represents the value matrix; d represents the feature dimension; softmax represents the soft maximization function; T represents the matrix transpose;

[0194] Here is the calculation When Q comes from the source domain Q=Z s , K, V come from the target domain features K=Z t , V = Z t ,calculate The opposite is true.

[0195] Specifically, in Step 33, four sets of feature flows (Z s ,Z t→s ,Z s→t ,Z t ) is passed to the next stage, but only the source dominant feature Z t→s For final inspection.

[0196] Specifically, Step 4~6: Repeat Step 3 to gradually deepen feature abstraction and alignment.

[0197] Specifically, in Step 7, based on low-level features and high-level features, the SSD unidirectional detector is used for training. At the same time, multi-level optimization is introduced in stages 2 (shallow layer), 4 (middle layer), and 6 (deep layer) to calculate the entropy loss of each stage (shallow layer, middle layer, deep layer): the negative response is suppressed by the MarginReLU function σm(x)=, and the cross-domain feature entropy is calculated:

[0198] ;

[0199] In the formula, represents the feature entropy of the lth layer from the target domain to the source domain; k represents the scaling factor (usually a positive constant); σ m represents the edge activation function (MarginReLU function is used in this embodiment); Represents the kth feature of the lth layer from the target domain to the source domain;

[0200] The total entropy loss is the average of each layer, which forces the model to reduce feature uncertainty and strengthen cross-domain consistency:

[0201] ;

[0202] Where, L entropy represents the total entropy loss; N represents the number of selected feature layers; l=2, 4, 6 represents the selected feature layer index; Represents the l-th layer feature entropy from the target domain to the source domain; represents the l-th layer feature entropy from the target domain to the source domain.

[0203] In one embodiment, combining transfer learning pre-trained weights and incremental learning framework to dynamically update model parameters and improve recognition performance of new categories includes:

[0204] S231, inputting the optimized source dominant features into the detection head for training, generating bounding box regression results and category scores;

[0205] S232, optimizing domain-invariant features through end-to-end training to achieve robust detection of the target domain;

[0206] S233, adopt incremental learning strategy to dynamically update model parameters, maintaining historical knowledge while adapting to new categories.

[0207] Specifically, Step 8: Add detection head training. Source dominant features output by the sixth stage The input is the detection head of the unidirectional detector (SSD) to generate the bounding box regression result Bpred and the category score Cpred. Through end-to-end training, the model gradually optimizes the domain-invariant features and finally achieves robust detection in the target domain. In each stage, the domain shift problem is systematically solved through fine-grained alignment of the spatial selective scanning mechanism (SSM), cross-domain interaction of the attention mechanism, and joint optimization of entropy and adversarial.

[0208] Specifically, after the classification device of the present invention is started, the display screen 501 will automatically play a video promoting garbage classification, which accounts for 3 / 4 of the video, and the right side is reserved for information to be classified. When garbage is put in, the V-type conveyor belt assembly starts to work, and the two first stepper motors 201 rotate at differential speeds, driving the synchronous belt 206 to rotate to complete garbage separation and transportation; when garbage enters the horizontal conveyor belt assembly from the V-type conveyor belt assembly, the V-type conveyor belt assembly immediately stops working and enters the detection stage.

[0209] Specifically, the camera 502 photographs the garbage on the horizontal conveyor belt assembly and transmits it to the development board for processing. It recognizes and classifies the photographed garbage photos by combining the hybrid domain adaptive method of Mamba-Transformer, and transmits the classification information to the control board. The control board controls the movement of the horizontal conveyor belt assembly, the compression unit 3 and the rotating storage unit 4 according to the information.

[0210] Specifically, according to the identified type of garbage, it is divided into: recyclable garbage, kitchen waste, hazardous waste and other garbage; if the identified type is recyclable garbage, the control panel controls the movement of the horizontal conveyor belt assembly, the sliding door 315 is closed, the electric push rod 301 is compressed, and the rotary storage unit 4 is rotated until the recyclable garbage bin coincides with the lower discharge port. After the compression is completed, the sliding door 315 opens to complete the classification; if the identified type is non-recyclable garbage, the control panel controls the movement of the horizontal conveyor belt, the sliding door 315 is opened, and the rotary storage unit 4 is rotated until the corresponding garbage bin coincides with the lower discharge port to complete the classification.

[0211] Specifically, after identification and classification are completed, the display screen automatically displays garbage classification information, including: garbage type, garbage quantity, whether the garbage has been disposed of, etc.; an infrared ranging sensor 105 is installed on the middle support plate 103 above the garbage can body 404, which is used to detect the loading status of the garbage can. When the garbage capacity exceeds 75%, the display screen 501 actively displays that the garbage can has reached a full load state, so that users can dump garbage in time.

[0212] In order to facilitate understanding of the above technical solution of the present invention, the following is a specific description in conjunction with embodiments as follows:

[0213] The garbage classification device proposed in the present invention works in coordination through a multi-stage transmission unit 2, a compression unit 3, a rotating storage unit 4 and an identification display unit 5. The linkage control between the units is realized by a control unit 6 composed of a development board (such as NVIDIA Jetson Nano development board) and a self-designed main control PCB board. The control unit 6 controls the working state of the conveyor belt, the electric push rod 301, the rotating chassis 403 and the display screen 501 according to the identified garbage type, realizes the function of simultaneously inputting multiple garbage into the classification, and improves the reliability and accuracy of the classification.

[0214] The multi-stage transmission unit 2 includes a V-shaped conveyor belt assembly and a horizontal conveyor belt assembly. The V-shaped conveyor belt assembly includes a left conveyor belt module and a right conveyor belt module, which have exactly the same structure and separate garbage through differential rotation to form a pair of transmission modules; the V-shaped conveyor belt assembly is composed of two differentially operated conveyor belts, which are differentially driven by two first stepper motors 201, and the angle between the conveyor belts is 110°~130°. The angle adjustment is achieved by an integrated stepper motor through a worm gear mechanism. The composite profile 202 used as a support frame is made of carbon fiber composite aluminum profile, the surface is coated with a polyurethane anti-slip layer, and an infrared density sensor (the model used in this embodiment is E3Z-T61) is integrated to detect the garbage stacking density in real time (the detection range is set to 0~200cm in this embodiment 2 ), when the density is ≥3 pieces / 100cm², the angle is automatically adjusted to 110°, and when the density is ≤1 piece / 100cm 2 When adjusting to 130°.

[0215] The compression unit 3 includes an electric push rod assembly and a sliding door assembly. The electric push rod 301 adopts a 24V / 1000N specification, which provides the force required for compression; the sliding door 315 is powered by the third stepper motor 310, which converts the rotational motion of the third stepper motor 310 into the linear motion of the sliding door, providing a compression platform for recyclable garbage; the overall frame of the device provides a fixed space for the electric push rod and the sliding door, improving the stability during compression.

[0216] The rotating storage unit 4 includes a rotating chassis 403 and a trash can body 404, which provide storage space for the classified trash; the rotating chassis 403 is powered by the fourth stepping motor 401, and the rotating chassis 403 is driven to rotate by the gear meshing transmission power, and a ball bearing is installed at the bottom of the rotating chassis 403 to reduce friction and provide support. The trash can body 404 adopts a cylindrical design to facilitate providing a larger capacity.

[0217] The identification and display unit 5 is composed of a camera 502 and a display screen 501, which provides reliability and interactivity of identification; the camera 502 adopts a 4K / USB wide-angle camera module to improve the detection range; the display screen 501 adopts a 1280*800 resolution to provide users with a high-quality interactive experience.

[0218] The V-belt assembly and the horizontal conveyor assembly provide reliable transportation functions; the two first stepper motors 201 realize closed-loop differential control through the control board, and the dynamic adjustment range of the differential ratio is 1:1.2~1:1.8, corresponding to the linear speed of the conveyor belt of 0.25m / s and 0.3~1.25m / s respectively. The control logic is based on the difference in garbage volume recognized by the camera (in this embodiment, the differential ratio is increased to 1:1.8 when it is set to >30%) and the friction coefficient of the material (such as the glass material is reduced to 1:1.3), combined with the pressure sensor (in this embodiment, the range is set to 0~50N, the accuracy is ±0.5%) to feedback the separation force in real time, and when the separation force is <5N, the bottom vibration motor 205 is triggered to assist in separation. The composite profile 202 used as the conveyor belt support frame is made of European standard 2060 carbon fiber composite aluminum profile (the composite profile model in this embodiment is CF-AL6061, the cross-sectional size is 20mm×60mm, and the bending strength is ≥200MPa), the surface of the conveyor belt is designed with fishbone anti-skid patterns (the depth of the fishbone anti-skid patterns in this embodiment is 1.2mm, the spacing is 8mm, and the inclination angle is 45°), and the bottom is integrated with a vibration motor 205 (the installation spacing between the vibration motor and the V-shaped conveyor belt assembly in this embodiment is 150mm). The conveyor belt is fixed to the European standard 2020 aluminum profile of the frame body 1 (the wall thickness of the frame body in this embodiment is 1.5mm) by M8 high-strength bolts (the bolt strength in this embodiment is 8.8 grade).

[0219] The sliding door 315 is driven by a third stepper motor 310 to move the gear rack to realize the opening and closing function of the lower release port; the electric push rod assembly is composed of an electric push rod 301 and a compression plate, wherein the compression plate is composed of an active compression plate 303 and a passive compression plate 308; the active compression plate 303 is connected to the electric push rod 301 by bolts, and is fixed to the left and right first optical axes 306 by a linear motion bearing 305 and an optical axis support seat 307, so that it has good stability; the passive compression plate 308 is fixed to the outer frame by a fiberglass board, so that the electric push rod 301 can adjust the compression space and compression capacity.

[0220] The rotating chassis 403 is powered by the fourth stepper motor 401, fixed and supported by bolts and balls, and saves bottom space; the chassis of the rotating storage unit 4 is composed of the rotating chassis 403 and the driven gear 406, both of which are radially positioned by bearings and the second optical axis 407, and axially positioned by balls, and the two cooperate to provide reliable positioning; the fourth stepper motor 401 drives the driving gear 402 to rotate, and the gear meshing drives the driven gear 406 to rotate, and the driven gear 406 is connected to the rotating chassis 403 by bolts, thereby driving the trash can body 404 to rotate, thereby realizing the classification of different garbage.

[0221] This device adopts a self-developed control board to complete the power supply and control of two first stepper motors 201, a second stepper motor 209, a third stepper motor 310, a fourth stepper motor 401, a display screen 501 and a development board; the camera 502 and the display screen 501 are connected to the development board, and the development board and the control board cooperate with each other to complete the garbage type identification and classification work; the power supply is completed by the Grignard 6S lithium battery model aircraft battery, and a photoelectric emergency stop sensor 7 is provided as a one-button start-stop switch to meet daily use requirements.

[0222] Finally, the garbage sorting device of the present invention is encapsulated by European standard 2020 aluminum profile, and a power switch is left on the side wall of the device, ensuring the versatility and adaptability of the device in a variety of application environments.

[0223] In summary, with the help of the above technical solution of the present invention, through the deep integration of hardware collaboration and intelligent algorithms, the industry pain points of low recognition accuracy and limited classification efficiency of traditional garbage sorting equipment in complex scenarios are solved. Compared with the existing technology, this device has the following innovative advantages:

[0224] 1) Dynamic separation-recognition collaborative architecture:

[0225] The cascade structure of V-type differential conveyor belt and horizontal conveyor belt is adopted, and the "separation-pause-recognition" working mechanism is pioneered. The physical dispersion of garbage is achieved through the differential control of two first-step motors 201, and the high-definition image acquisition is completed in a static state with the camera, which significantly improves the quality of feature extraction. According to the test, this design improves the accuracy of small-target garbage recognition by 37%, effectively overcoming the image blur problem caused by traditional continuous transmission.

[0226] 2) Intelligent compression-storage linkage system:

[0227] The electric push rod compression module is innovatively dynamically coupled with the rotating storage unit 4, and millimeter-level timing control is achieved through the control panel. When it is identified as recyclables, the system completes the linkage operations of sliding door closing, compression execution (pressure up to 200N), and storage bin positioning within 0.5 seconds. The compression efficiency is 60% higher than that of traditional pneumatic solutions, and energy consumption is reduced by 45%.

[0228] 3) Multi-dimensional state-aware network:

[0229] A multimodal monitoring system consisting of infrared ranging sensors, motor torque detection, and visual feedback was built to sense the capacity of the trash can (accuracy ±2mm), the operating status of the equipment, and abnormal vibration in real time. An LSTM prediction model was established through the development board, which can warn of equipment failures 10 minutes in advance and shorten the operation and maintenance response time by 80%.

[0230] 4) Adaptive learning framework:

[0231] The device has a built-in incremental learning algorithm that uses tens of thousands of garbage image data processed daily to continuously optimize the YOLOv8 model through a comparative learning strategy. Actual deployment data shows that weekly model iterations of the system can increase the accuracy of identifying new types of garbage by 15%, effectively solving the performance degradation problem of traditional fixed models caused by changes in garbage morphology.

[0232] The device has been verified by laboratory simulation and field testing, achieving a classification accuracy of 95.3% in a test set containing more than 200 types of garbage, and a processing speed of 120 pieces / minute. Compact modular design (occupying an area of ​​<1.5m 2 ) with a universal ball mechanism, it is particularly suitable for space-constrained scenarios such as schools and communities, providing an efficient and reliable garbage sorting solution for smart city construction.

[0233] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A self-supervised hybrid domain adaptive multi-waste classification device, characterized in that: It includes a frame body, a multi-stage transmission unit, a compression unit, a rotation storage unit and an identification display unit, wherein the multi-stage transmission unit, the compression unit and the rotation storage unit are sequentially arranged inside the frame body from top to bottom, and the multi-stage transmission unit is located on one side of the central axis of the frame body, and the compression unit is located on the other side of the central axis of the frame body; A delivery port is provided at the top of the frame body, which is located above the multi-stage conveying unit. Ball bearings are fixedly provided at the four corners of the bottom end of the frame body. A middle support plate is provided in the middle of the frame body. A lower base plate is provided at the bottom of the frame body. A control unit is provided on one side of the frame body.

2. The self-supervised hybrid domain adaptive multi-waste classification device according to claim 1, characterized in that: The multi-stage conveying unit includes a V-shaped conveyor belt assembly arranged on one side of the top of the frame body; The V-shaped conveyor belt assembly includes two groups of symmetrically arranged composite profiles, and the two groups of composite profiles are in a V-shaped obtuse structure, and the bottom ends of the two groups of composite profiles are provided with a plurality of linearly distributed V-shaped conveyor belt connectors, and one side of the bottom of the two groups of composite profiles is commonly connected to a vibration motor; Both ends of the composite profile are provided with V-shaped conveyor belt rollers, and one end of the two groups of V-shaped conveyor belt rollers located on one side of the vibration motor is provided with a first stepper motor.

3. The self-supervised hybrid domain adaptive multi-waste classification device according to claim 2, characterized in that: The multi-stage conveying unit further comprises a horizontal conveyor belt assembly arranged in the middle of the frame body on one side away from the V-shaped conveyor belt assembly; The horizontal conveyor belt assembly comprises two groups of symmetrically arranged horizontal conveyor belt rollers, the outer sides of the horizontal conveyor belt rollers are sleeved with horizontal conveyor belts, both ends of the horizontal conveyor belt rollers are provided with bearing seats, and a connecting profile is provided between the two groups of the bearing seats; A second stepper motor is arranged at the bottom end of the connecting profile, and a synchronous wheel is sleeved on the output end of the second stepper motor. The synchronous wheel is connected to a horizontal conveyor belt roller away from the compression unit side through a synchronous belt.

4. The self-supervised hybrid domain adaptive multi-waste classification device according to claim 1, characterized in that: The compression unit includes an electric push rod assembly arranged on one side of the middle part of the frame body; The electric push rod assembly includes a rear support plate arranged on one side of the middle part of the frame body, an electric push rod is arranged on one side of the rear support plate, an active compression plate is arranged at the output end of the electric push rod, a first optical axis is penetrated at both ends of the active compression plate, an optical axis support frame is arranged at one end of the first optical axis, and a linear motion bearing fixedly connected to the active compression plate is arranged in the middle of the first optical axis; A front support plate is commonly provided at the other end of the two groups of the first optical axes, and optical axis support seats matching the first optical axes are provided on both sides of the front support plate, and a passive compression plate is provided in the middle of the front support plate.

5. The self-supervised hybrid domain adaptive multi-waste classification device according to claim 4, characterized in that: The compression unit further comprises a sliding door assembly arranged in the middle of the frame body and below the electric push rod assembly; The sliding door assembly comprises two sets of sliding door tracks symmetrically arranged at the top of the middle support plate, one side of the sliding door track is provided with a sliding groove, and the inner side of the sliding groove is provided with a sliding door; A sliding door upper rack is arranged on one side of the top end of the sliding door, and a third stepper motor is arranged on the side of the sliding door track away from the multi-stage transmission unit. The output end of the third stepper motor is arranged with a sliding door gear matched with the sliding door upper rack.

6. The self-supervised hybrid domain adaptive multi-waste classification device according to claim 1, characterized in that: The rotating storage unit comprises a rotating chassis arranged at the bottom of the frame body, a plurality of trash can bases distributed in a circumference are opened at the top of the rotating chassis, a trash can body is arranged on the top of the trash can base, and a driven gear is arranged at the bottom of the rotating chassis; a fourth stepping motor is arranged at one side of the rotating chassis below the compression unit, and a driving gear meshing with the driven gear is arranged at the bottom of the output end of the fourth stepping motor; An optical axis fastener is provided through the middle of the rotating chassis, a second optical axis is provided through the middle of the optical axis fastener, a middle fixing piece matching with the optical axis fastener is sleeved on the outer side of the middle of the second optical axis, tapered roller bearings are sleeved on the top and bottom ends of the second optical axis, and a fixing piece is provided on the outer side of the tapered roller bearing.

7. A self-supervised hybrid domain adaptive multi-waste classification control method, using the self-supervised hybrid domain adaptive multi-waste classification device according to any one of claims 1 to 6 to realize multi-waste classification control, characterized in that: include: SI, separate the input garbage through the V-type conveyor belt assembly at a differential speed, and dynamically adjust the differential ratio through the control unit; SII, pause the horizontal conveyor assembly, use the camera to collect a static image and transmit it to the control unit; SIII, the control unit completes the garbage classification identification and outputs the result to the control board when the confidence reaches the threshold; SIV, select the operation mode according to the classification result. When it is identified as recyclable garbage, execute the recyclable garbage mode, and execute the operations of closing the sliding door, compressing the electric push rod, rotating the rotating chassis and opening the sliding door in sequence; when it is identified as non-recyclable garbage, execute the non-recyclable garbage mode, and execute the operations of rotating the rotating chassis and opening the sliding door in sequence; SV, updates the garbage classification information and capacity status through the display screen, and triggers a shutdown warning when the capacity of the garbage bin exceeds the threshold.

8. A method for multi-waste classification and identification using a self-supervised hybrid domain adaptive multi-waste classification device according to any one of claims 1 to 6, characterized in that: include: S1, obtain multi-view images of garbage through the camera and pre-process them; S2. Use the hybrid domain adaptation garbage recognition model to extract image features, realize spatial selective scanning and channel switching, and use the transfer learning and incremental learning framework to update the model parameters; S3, based on the hybrid domain adaptive garbage identification model, output the garbage classification results and confidence, and synchronously update them to the display screen and control unit; Wherein, the S2 includes: S21. Build a hybrid domain adaptation garbage recognition model to improve feature extraction capabilities by embedding channel attention modules and contrastive learning loss functions; S22, processing feature maps based on spatial selective scanning mechanism and channel exchange strategy, and realizing spatial selective scanning and channel exchange by calculating cross-domain feature entropy; S23. Combine the transfer learning pre-trained weights and incremental learning framework to dynamically update the model parameters and improve the recognition performance of garbage categories.

9. The method for multi-waste classification and identification based on self-supervised hybrid domain adaptation according to claim 8, characterized in that: The hybrid domain adaptation garbage recognition model is constructed to improve the feature extraction capability by embedding the channel attention module and the contrastive learning loss function, including: S211, extracting shallow features of source domain and target domain images through two consecutive convolution blocks, and generating preliminary feature maps through maximum pooling downsampling; S212, using depthwise separable convolution to process the preliminary feature map, measuring the spatial consistency of the source domain feature and the target domain feature map by cosine similarity, and generating a similarity matrix; S213, equally dividing the source domain feature map and the target domain feature map along the channel dimension, calculating the channel-level similarity after exchanging some channels, and generating features that are spatially aligned with the channels; The expression of the channel-level similarity is: ； In the formula, S channel represents channel-level similarity; [ , ] represents the feature vector of the concatenation of segments 1 and 2 of the source domain feature map and segments 3 and 4 of the target domain feature map; [ , ] represents the feature vector obtained by concatenating segments 1 and 2 of the target domain feature map and segments 3 and 4 of the source domain feature map; ||·|| represents the norm of the vector.

10. The method for multi-waste classification and identification based on self-supervised hybrid domain adaptation according to claim 8, characterized in that: The processing of feature maps based on the spatial selective scanning mechanism and the channel exchange strategy, and the implementation of spatial selective scanning and channel exchange by calculating cross-domain feature entropy include: S221, using self-attention to independently enhance the intra-domain features, generating source dominant features and target dominant features through cross-attention, and achieving soft alignment of features; S222, passing the source domain features, the source dominant features, the target dominant features, and the target domain features to the next stage, and repeating the feature extraction and alignment process; S223, introduce entropy loss in shallow, middle and deep layers, calculate cross-domain feature entropy through edge activation function, and strengthen cross-domain consistency of multi-level features; The expression of the cross-domain feature entropy is: ； ； In the formula, H t→ts (l) represents the entropy value of the dominant characteristic of the source in the first stage; k represents the proportionality coefficient; σ m represents the edge activation function; Z t→ts (l,k) Indicates the dominant characteristic of the source in stage l; Z t (1,k) represents the target domain features of the first stage; L entropy represents the total entropy loss; N represents the number of feature layers; l represents the feature layer index; H t→st (l) Represents the entropy value of the target dominant feature in the lth stage.

Images