A garbage collection method and system based on multi-dimensional vision and multi-dimensional weight estimation

Abstract

This invention discloses a waste recycling method and system based on multi-dimensional vision and multi-dimensional weight estimation, comprising: acquiring visual images of the waste and fusing them with three-dimensional morphology to identify the type of the waste; acquiring total weight and compacted volume data, and estimating the weight of cardboard boxes, metals, and textiles by combining the type of waste with value weighting coefficients and density weighting coefficients to match different values ​​for each category and to price them accordingly, while simultaneously making a reasonableness judgment to prevent fraud; after determining compliance, generating an account review request and temporarily storing it on an edge computing chip when the network is offline, waiting for the network to be restored and then simultaneously generating user points settlement data on the blockchain; after cloud-based data acquisition and review, the points are credited to the user's account for consumption or withdrawal. Its beneficial effects are that it achieves accurate differentiated value pricing and multi-dimensional anti-fraud verification, and improves the reliability of settlement and the security of points withdrawal in offline environments.

Description

Technical Field

[0001] This invention relates to the field of smart sanitation, specifically to a waste recycling method and system based on multidimensional vision and multidimensional weight estimation. Background Technology

[0002] Current waste recycling technologies typically rely on simple single vision or sensors to identify the type of items thrown in, resulting in low accuracy in identifying material composition and susceptibility to contamination. Regarding pricing and settlement, existing technologies only perform simple overall weight calculations based on the detected item type. Since the value of recyclable cardboard boxes, metals, and textiles differs, simple pricing fails to reflect the true economic value of each category. Furthermore, current technologies do not comprehensively estimate and reasonably calculate the weight of cardboard boxes, metals, and textiles by combining item type, total weight, compacted volume, and corresponding value and density weight coefficients for each item category. This leads to large pricing errors and makes it easy for heavy items to be added to accumulate points. Moreover, the user's point verification and consumption mechanism is imperfect, lacking a complete data loop to prevent tampering, making it difficult to ensure the accuracy and security of account settlement and point display. Summary of the Invention

[0003] This invention proposes a waste recycling method based on multidimensional vision and multidimensional weight estimation, comprising: S1. The visual image data and three-dimensional morphology data of the delivered object are acquired by the multi-dimensional sensing acquisition device, and the delivered object is visually detected and feature fusion recognized to obtain the item type recognition result. S2. The total weight data and compacted volume data of the delivered item are obtained by the weight acquisition device and the volume calculation device. Based on the item type identification result, the total weight data and the compacted volume data, the weight of the cardboard box, metal and textile is estimated by the value weight coefficient and the density weight coefficient and priced accordingly. At the same time, a reasonableness logic judgment is performed to prevent cheating and a verified pricing result is obtained. S3. When the verification pricing result is determined to be compliant, an account review request is generated based on the verification pricing result. In the offline state, the account review request is temporarily stored in the local ledger of the edge computing chip and the blockchain light node. After the network is restored, it is uploaded to the chain in batches and synchronized to generate user points settlement data. S4. The cloud server obtains the user points settlement data, performs account verification on the user points settlement data, and displays the verified points data to the user account for points consumption or points withdrawal.

[0004] Furthermore, the multi-dimensional sensing acquisition device includes a multispectral vision camera and a millimeter-wave radar. The multi-dimensional sensing acquisition device constructs the three-dimensional shape and visual material features of the delivered item when it falls, and outputs the item type identification result of cardboard box, metal or textile through edge computing chip fusion processing. The item type identification result is the category label of cardboard box, metal or textile output after the multi-spectral visual and three-dimensional morphological fusion analysis of the delivered item is performed by the multi-dimensional sensing acquisition device; The verification pricing result is a weight estimation conclusion, a pricing amount conclusion, and a reasonableness judgment conclusion calculated and output by the edge computing chip based on the item type identification result, the value weight coefficient and the density weight coefficient, and in combination with the total weight data and the compacted volume data. The user points settlement data is an immutable points transfer record generated on the blockchain after the network is restored, based on the compliant account review request. The user account is a digital asset account that receives approved points data and provides consumption or withdrawal operations.

[0005] Furthermore, the weight estimation formula is as follows: The pricing formula is The reasonableness determination formula is as follows: ,like If it is determined to be cheating, then, Representing the Estimated weight of this type of item This represents the total weight data. represent The median of the standard density range for this type of item Representing the Volume data of the type of item, Represents the sum of all identified item categories. Represents the pricing amount. Representing the The value weighting coefficient of the category of items, Represents the rationality ratio coefficient. The median of the composite standard density representing the mixed delivery materials, This represents the volume data after compaction. The lower threshold representing the reasonableness ratio range. The upper limit threshold represents the range of reasonableness ratios.

[0006] Furthermore, when the network is down, the device switches to local island mode and can still complete visual recognition, weighing and weight estimation, and reasonableness verification. After the network is restored, the double-blind hash data during the offline period will be uploaded to the blockchain in batches to ensure that the distribution of points cannot be tampered with.

[0007] Furthermore, all delivery and settlement data within the area are aggregated to the regional dispatch center to predict future waste increase curves, automatically generate collection and transportation needs, and distribute routes and types of waste to be loaded at each station to the sanitation vehicle fleet. After the vehicles arrive, they are automatically checked and opened via smart locks, achieving a balance of data ledgers across the three ends.

[0008] Furthermore, the cloud server dynamically allocates visual inspection threads, weighing platform operating frequency, and pneumatic nozzle operating sequence and energy consumption limits based on the digital twin sand table.

[0009] Furthermore, the multispectral vision camera captures visible light and infrared spectral images, and the millimeter-wave radar captures three-dimensional contour point clouds. The data from both are fused at the feature level through a lightweight AI large model within the edge computing chip, thereby outputting the item type recognition result.

[0010] Furthermore, the present invention also proposes a non-volatile storage medium storing a computer program, which, when executed by a processor, implements the aforementioned garbage collection method based on multidimensional vision and multidimensional weight estimation.

[0011] Furthermore, this invention also proposes a waste recycling system based on multidimensional vision and multidimensional weight estimation, comprising: The identification module is used to process the results of the delivery item type identification. The multidimensional estimation and verification module is used to process the verification pricing results obtained from multidimensional data collection. The points settlement module is used to process user points settlement data from compliance judgment. The account withdrawal module is used to process user points settlement data into user account points withdrawal.

[0012] This invention proposes a waste recycling method based on multidimensional vision and multidimensional weight estimation. It utilizes a multidimensional sensing acquisition device to achieve high-precision visual detection and identification of items such as cardboard boxes, metals, and textiles. By combining item type, total weight, compacted volume, value weight coefficient, and density weight coefficient, the weight of cardboard boxes, metals, and textiles is multidimensionally estimated to adapt to the different recycling values ​​of each category and to correctly price them. Simultaneously, a consistency verification formula is used to calculate the reasonableness of items to prevent cheating. Localized anti-cheating verification and blockchain light nodes are deployed on edge computing chips, effectively alleviating the problems of high false positive rates and the issue of heavy items being used to accumulate points. This significantly improves the reliability of settlement in offline environments and the security of user account points, achieving a closed-loop data system with end-to-end anti-tampering. Attached Figure Description

[0013] Figure 1 This is the overall logic execution topology diagram of the method of the present invention.

[0014] Figure 2 This is a diagram showing the internal data flow topology of the multidimensional weight estimation and pricing verification module of this invention.

[0015] Figure 3 This is the control topology diagram for the offline island mode and on-chain synchronization of the present invention.

[0016] Figure 4 This is the scheduling topology diagram for the regional scheduling and pre-positioning of logistics teams for waste removal in this invention.

[0017] Figure 5 This is the allocation topology diagram for the cloud-based digital twin energy consumption management system of this invention.

[0018] Figure 6 This is a schematic diagram of the process of a waste recycling method based on multidimensional vision and multidimensional weight estimation according to the present invention. Detailed Implementation

[0019] This invention proposes a waste recycling method based on multidimensional vision and multidimensional weight estimation, comprising: See Figure 1 and Figure 6 The overall logic execution topology diagram of the method of this invention shows the entire data flow and feedback mechanism from the falling of the delivered object to the final integration.

[0020] S1. The visual image data and three-dimensional morphological data of the delivered object are acquired by the multi-dimensional sensing acquisition device, and the delivered object is visually detected and feature fusion is performed to obtain the item type identification result.

[0021] The item type identification result is a label indicating whether the item is cardboard, metal, or textile, output after the multi-spectral visual and three-dimensional morphological fusion analysis is performed on the disposed item by the multi-dimensional sensing acquisition device. The disposed item refers to the physical waste falling through the disposal port. The multi-dimensional sensing acquisition device is a combination of a multi-spectral visual camera and millimeter-wave radar for multi-dimensional data acquisition. When the waste falls, the multi-spectral visual camera captures visible light and infrared spectral images, and the millimeter-wave radar captures three-dimensional contour point clouds. The sensors construct its three-dimensional morphology and visual material features within a very short time window, distinguishing between cardboard, metal, and textile categories. After fusion processing by an edge computing chip, the item type identification result is output. When a user disposes of mixed waste, the multi-spectral visual camera identifies the reflectance spectrum of the cardboard fiber material and the metallic luster, the millimeter-wave radar constructs its respective square and irregular three-dimensional contours, and the edge computing chip fuses and determines that it contains both cardboard and metal.

[0022] Furthermore, the multi-dimensional sensing acquisition device includes a multispectral vision camera and a millimeter-wave radar. The multi-dimensional sensing acquisition device constructs the three-dimensional shape and visual material features of the delivered object as it falls, and outputs the object type identification result (cardboard box, metal, or textile) through edge computing chip fusion processing.

[0023] Among them, the multispectral vision camera is a visual sensor that captures multi-band spectral reflectance. The millimeter-wave radar is a ranging sensor that emits high-frequency electromagnetic waves to construct a 3D point cloud. The multispectral vision camera is responsible for capturing surface texture and material properties, while the millimeter-wave radar is responsible for capturing the 3D geometric contours. The data from both are fused at the feature level within an edge computing chip using a lightweight AI large model. When a plastic bottle covered in mud is placed in the machine, a single vision camera struggles to identify the material. However, the contours constructed by the millimeter-wave radar and the multispectral infrared reflectance are fused using a large model, successfully outputting the plastic product identification result.

[0024] Furthermore, the multispectral vision camera captures visible light and infrared spectral images, and the millimeter-wave radar captures three-dimensional contour point clouds. The data from both are fused at the feature level through a lightweight AI large model within the edge computing chip, thereby outputting the item type recognition result.

[0025] Feature-level fusion is a data processing procedure that involves concatenating and weighting feature vectors extracted from multiple sensors in the hidden layer of the model. This multimodal feature fusion mechanism effectively overcomes the blind spots in recognition by a single sensor when it is soiled or obstructed. When a wet wad of paper is placed in the sensor, a multispectral vision camera identifies its moisture content and fiber characteristics, while a millimeter-wave radar constructs its irregular wad shape; the fusion process then outputs the correct type.

[0026] S2. The total weight data and compacted volume data of the delivered item are obtained by the weight acquisition device and the volume measurement device. Based on the item type identification result, the total weight data and the compacted volume data, the weight of the cardboard box, metal and textile is estimated by the value weight coefficient and the density weight coefficient and priced accordingly. At the same time, a reasonableness logic judgment is performed to prevent cheating, and a verified pricing result is obtained.

[0027] The verification pricing result is a weight estimation conclusion, pricing amount conclusion, and reasonableness judgment conclusion calculated and output by the edge computing chip based on the item type identification result, retrieving the value weight coefficient and the density weight coefficient, and combining the total weight data and the compacted volume data. The total weight data is a physical mass measurement value collected by a weight acquisition device. The compacted volume data is a three-dimensional total volume measurement value of the space after acoustic fluidization rearrangement compaction, calculated by a volume measurement device. The value weight coefficient refers to a pre-set unit price coefficient based on the real-time economic value of each category of items in the recycling market. The density weight coefficient refers to the proportion of the standard physical density reference value of each category of items in the total volume estimation of the mixed submission. (See...) Figure 2 The internal data flow topology diagram of the multi-dimensional weight estimation and pricing verification module shows a dual closed-loop verification mechanism that derives the total weight from the volume and density of various individual items, and then derives the pricing amount by combining value weights. The edge computing chip retrieves the standard density range and value weight coefficients for each category based on the item type identification result. Since the value of the cardboard boxes, metals, and textiles to be recycled is different, the system substitutes the standard density range, the total weight data, the compacted volume data, and the density weight coefficients into the weight estimation formula and the reasonableness judgment formula to calculate and output the verified pricing result. The weight estimation formula is as follows: The pricing formula is The reasonableness determination formula is as follows: ,like If it is determined to be cheating, then, Representing the Estimated weight of this type of item This represents the total weight data. represent The median of the standard density range for this type of item Representing the Volume data of the type of item, Represents the sum of all identified item categories. Represents the pricing amount. Representing the The value weighting coefficient of the category of items, Represents the rationality ratio coefficient. The median of the composite standard density representing the mixed delivery materials, This represents the volume data after compaction. The lower threshold representing the reasonableness ratio range. The upper threshold representing the reasonableness ratio range. (See Table 1) Table of Item Value Weights and Density Parameters: Table 1

[0028] For example, the total weight invested by the user is The volume of the cardboard box and metal mixture, after compaction, was calculated by a volume measuring device to be: The volume of the cardboard box was calculated using visual recognition. The volume of the metal is The system retrieves the median standard density of the cardboard box. The median standard density of metals is Substitute the values ​​into the weight estimation formula to obtain the estimated weight of the cardboard box and the estimated weight of the metal; then, based on the value weight of the cardboard box... Metal value weight Substitute into the pricing formula to calculate the pricing amount. Simultaneously, calculate the rationality ratio coefficient. It was found to be beyond The specified range indicates a discrepancy in physical density between the carton ratio and the metal ratio. The system determines this as fraud and refuses billing, adding the information to the device's local blacklist. (See Table 2) Reasonableness Judgment and Fraud Prevention Logic Table: Table 2

[0029] Furthermore, the edge computing chip retrieves the standard density range and the value weight coefficient of the corresponding item based on the item type identification result, substitutes the standard density range, the total weight data, the compacted volume data, and the density weight coefficient into the weight estimation formula and the reasonableness judgment formula to calculate and output the verification pricing result.

[0030] The standard density range is a pre-set reference range of physical density for various common types of waste in the local database. The value weighting coefficient is a pre-set unit price coefficient due to the different recycling values ​​of each category of items. The system extrapolates the theoretical weight of each category in the total weight by using the volume and standard density of multiple categories, thereby restoring the true weight of each item for fair pricing. For example, if the submitted item consists of cardboard boxes and textiles, since textiles are denser than cardboard boxes, the system automatically and reasonably allocates the total weight to cardboard boxes and textiles based on the product of volume and standard density, avoiding value loss caused by uniform pricing.

[0031] Furthermore, the weight estimation formula is as follows: The pricing formula is The reasonableness determination formula is as follows: ,like If it is determined to be cheating, then, Representing the Estimated weight of this type of item This represents the total weight data. represent The median of the standard density range for this type of item Representing the Volume data of the type of item, Represents the sum of all identified item categories. Represents the pricing amount. Representing the The value weighting coefficient of the category of items, Represents the rationality ratio coefficient. The median of the composite standard density representing the mixed delivery materials, This represents the volume data after compaction. The lower threshold representing the reasonableness ratio range. This represents the upper threshold of the reasonableness ratio range. Reasonableness ratio coefficient. It is a dimensionless numerical value that measures the consistency of the macroscopic physical state of the delivered item. The system first estimates the weight through microscopic data of each category for refined pricing, and then uses the ratio of macroscopic overall data to prevent fraud, forming a double data verification closed loop. For example, the total volume of the delivered item may shrink significantly after being compacted by sound waves, but the total weight remains unchanged. The system calculates the density using macroscopic formulas to determine that it is reasonable, and then estimates the weight of cardboard boxes, metal, and textiles separately using microscopic formulas to achieve accurate pricing.

[0032] S3. When the verification pricing result is determined to be compliant, an account review request is generated based on the verification pricing result. In the offline state, the account review request is temporarily stored in the local ledger of the edge computing chip and the blockchain light node. After the network is restored, it is uploaded to the chain in batches and synchronized to generate user points settlement data.

[0033] The user points settlement data is an immutable points transfer record generated synchronously on the blockchain after network recovery based on the compliant account verification request. The account verification request is an encrypted data packet containing a summary of delivery characteristics and the pricing result. The edge computing chip is a hardware carrier deployed on the device for local data processing. (See...) Figure 3 The control topology diagram for the offline isolated mode and on-chain synchronization demonstrates the asynchronous processing flow of local autonomous settlement during network outages and batch on-chain processing after network recovery. When the network is down, the device switches to local isolated mode, still capable of visual recognition, weighing and weight estimation, and reasonableness verification. Lightweight AI models and blockchain light nodes are deployed directly on the edge computing chip at the device end. Once the network is restored, the double-blind hash data accumulated during the offline period is batch-uploaded and synchronized to the blockchain, ensuring the immutability of points distribution. For example, when a user submits compliant textiles during an offline state, the edge computing chip retrieves the local model to complete the estimation and pricing, generates an account verification request, and temporarily stores it in the chip's local ledger. Once the network is restored, the data is batch-uploaded to the blockchain, generating user points settlement data.

[0034] Furthermore, when the network is down, the device switches to local island mode and can still complete visual recognition, weighing and weight estimation, and reasonableness verification. After the network is restored, the double-blind hash data during the offline period will be uploaded to the blockchain in batches to ensure that the distribution of points cannot be tampered with.

[0035] The local island mode is a fully offline, autonomous operating state activated by the device when it senses a network outage. Double-blind hash data consists of unpredictable encrypted transaction records generated during the network outage. A lightweight AI model ensures sufficient computing power for offline identification, while lightweight blockchain nodes guarantee the reliable temporary storage and subsequent synchronization of offline transaction data onto the blockchain. For example, in an underground parking garage with a network outage, the device continues to receive deliveries and uses its local AI model to complete classification, weighing, and formula verification. The generated points transactions are packaged into hashes for temporary storage and automatically uploaded to the blockchain in batches once the network is restored.

[0036] S4. The cloud server obtains the user points settlement data, performs account verification on the user points settlement data, and displays the verified points data to the user account for points consumption or points withdrawal.

[0037] The user account is a digital asset account that receives approved points data and provides the ability to make purchases or withdraw points. The cloud server is a centralized processing cluster that aggregates global data and manages accounts. After receiving the user points settlement data synchronized to the blockchain, the cloud server verifies the user's identity information and the compliance of the delivery. Once approved, the corresponding points are credited to the user's associated account, which the user can then use to make purchases or withdraw points through compliant channels. For example, if a user delivers a metal aluminum can compliantly and completes the estimated pricing, the cloud server credits the corresponding points to the user's account, and the user can then make a purchase at a partner convenience store by scanning a QR code.

[0038] Furthermore, all delivery and settlement data within the area are aggregated to the regional dispatch center to predict future waste increase curves, automatically generate collection and transportation needs, and distribute routes and types of waste to be loaded at each station to the sanitation vehicle fleet. After the vehicles arrive, they are automatically checked and opened via smart locks, achieving a balance of data ledgers across the three ends.

[0039] Among them, the three-terminal data ledger balance refers to the logically absolute consistency and immutability of the ledger status of front-end delivery points settlement data, mid-end collection data, and end-of-line processing data. The regional dispatch center is a cloud or edge server cluster that aggregates digital twin data for the region. (See also...) Figure 4 The scheduling topology diagram for regional dispatching and pre-positioned logistics fleets for waste removal illustrates the entire process from summarizing and predicting incremental data from different areas to issuing waste removal instructions to the fleets, and finally balancing the data ledger. (See Table 3) Three-terminal data ledger balance verification table: Table 3

[0040] The system not only monitors point settlement but also predicts future incremental curves, deeply integrating data with the sanitation vehicle fleet's GPS and the physical partitions of the truck compartments. Before the collection trucks depart, the system distributes the routes and the types of waste to be loaded at each station to the vehicles. For example, if the dispatch center predicts that station A's cardboard boxes are full and the incremental load will reach saturation in the next few hours, it automatically issues a collection instruction to the fleet. The vehicles arrive according to the predetermined route, and after the smart locks verify and match, the doors open, and the corresponding cardboard boxes for station A are loaded as instructed.

[0041] Furthermore, the cloud server dynamically allocates visual inspection threads, weighing platform operating frequency, and pneumatic nozzle operating sequence and energy consumption limits based on the digital twin sand table.

[0042] Among them, the energy consumption limit is an upper limit indicator of equipment operating power calculated and issued by the dispatch center based on the current state of waste accumulation and sorting tasks. (See also) Figure 5 The cloud-based digital twin energy consumption management topology diagram illustrates the control logic for allocating working frequencies and timings to various terminals based on the digital twin sandbox in the cloud. By coordinating the timing of the vision and aerodynamic field generators of all devices within the area in the cloud, grid fluctuations caused by multiple devices operating at high frequencies simultaneously are avoided. For example, if the dispatch center detects that adjacent buckets B and C both require high-frequency vision detection and aerodynamic sorting at the same time, to avoid instantaneous current overload caused by simultaneous startup, bucket B is instructed to execute first, and bucket C's execution is delayed, achieving balanced energy consumption scheduling.

[0043] This invention also proposes a non-volatile storage medium storing a computer program. When executed by a processor, the computer program implements the garbage collection method based on multi-dimensional vision and multi-dimensional weight estimation as described in any of steps S1 to S4. The storage medium may include ROM, magnetic tape, hard disk, optical disk, or other carriers capable of persistently storing data. When the processor executes the program, it performs the data processing and logical judgments described in the above steps.

[0044] This invention also proposes a waste recycling system based on multidimensional vision and multidimensional weight estimation, comprising: an acquisition and identification module for processing the identification results from the deposited items to the item type; a multidimensional estimation and verification module for processing the verification and pricing results from multidimensional data collection; a points settlement module for processing the compliance judgment and user points settlement data; and an account display module for processing the user points settlement data and displaying the user's points in the user's account. The acquisition and identification module corresponds to the multidimensional vision and radar data parsing in S1; the multidimensional estimation and verification module corresponds to the weight estimation formula and anti-fraud formula calculation in S2; the points settlement module corresponds to the offline temporary storage and on-chain synchronization in S3; and the account display module corresponds to the cloud review and points distribution in S4. The data interaction relationships between the modules strictly correspond one-to-one with the input-output closed loop of weight 1.

[0045] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A waste recycling method based on multidimensional vision and multidimensional weight estimation, characterized in that, include: S1. The visual image data and three-dimensional morphology data of the delivered object are acquired by the multi-dimensional sensing acquisition device, and the delivered object is visually detected and feature fusion recognized to obtain the item type recognition result. S2. The total weight data and compacted volume data of the delivered item are obtained by the weight acquisition device and the volume calculation device. Based on the item type identification result, the total weight data and the compacted volume data, the weight of the cardboard box, metal and textile is estimated by the value weight coefficient and the density weight coefficient and priced accordingly. At the same time, a reasonableness logic judgment is performed to prevent cheating and a verified pricing result is obtained. S3. When the verification pricing result is determined to be compliant, an account review request is generated based on the verification pricing result. In the offline state, the account review request is temporarily stored in the local ledger of the edge computing chip and the blockchain light node. After the network is restored, it is uploaded to the chain in batches and synchronized to generate user points settlement data. S4. The cloud server obtains the user points settlement data, performs account verification on the user points settlement data, and displays the verified points data to the user account for points consumption or points withdrawal.

2. The waste recycling method based on multidimensional vision and multidimensional weight estimation as described in claim 1, characterized in that, The multi-dimensional sensing acquisition device includes a multispectral vision camera and a millimeter-wave radar. When the delivered item falls, the multi-dimensional sensing acquisition device constructs its three-dimensional shape and visual material features, and through edge computing chip fusion processing, outputs the item type identification result of cardboard box, metal or textile. The item type identification result is the category label of cardboard box, metal or textile output after the multi-spectral visual and three-dimensional morphological fusion analysis of the delivered item is performed by the multi-dimensional sensing acquisition device; The verification pricing result is a weight estimation conclusion, a pricing amount conclusion, and a reasonableness judgment conclusion calculated and output by the edge computing chip based on the item type identification result, the value weight coefficient and the density weight coefficient, and in combination with the total weight data and the compacted volume data. The user points settlement data is an immutable points transfer record generated on the blockchain after the network is restored, based on the compliant account review request. The user account is a digital asset account that receives approved points data and provides consumption or withdrawal operations.

3. The waste recycling method based on multidimensional vision and multidimensional weight estimation as described in claim 1, characterized in that, The edge computing chip retrieves the standard density range and value weight coefficient of the corresponding item based on the item type identification result. It then substitutes the standard density range, the total weight data, the compacted volume data, and the density weight coefficient into the weight estimation formula and the reasonableness judgment formula to calculate and output the verification pricing result.

4. The waste recycling method based on multidimensional vision and multidimensional weight estimation as described in claim 3, characterized in that, The weight estimation formula is as follows: The pricing formula is The reasonableness determination formula is as follows: ,like If it is determined to be cheating, then, Representing the Estimated weight of this type of item This represents the total weight data. represent The median of the standard density range for this type of item Representing the Volume data of the type of item, Represents the sum of all identified item categories. Represents the pricing amount. Representing the The value weighting coefficient of the category of items, Represents the rationality ratio coefficient. The median of the composite standard density representing the mixed delivery materials, This represents the volume data after compaction. The lower threshold representing the reasonableness ratio range. The upper limit threshold represents the range of reasonableness ratios.

5. The waste recycling method based on multidimensional vision and multidimensional weight estimation as described in claim 1, characterized in that, When the network is down, the device switches to local island mode and can still complete visual recognition, weighing and weight estimation and reasonableness verification. After the network is restored, the double-blind hash data during the offline period will be uploaded to the blockchain in batches to ensure that the points distribution cannot be tampered with.

6. The waste recycling method based on multidimensional vision and multidimensional weight estimation as described in claim 5, characterized in that, All delivery and settlement data within the area are aggregated to the regional dispatch center to predict future waste increase curves, automatically generate collection and transportation needs, and distribute routes and types of waste to be loaded at each station to the sanitation vehicle fleet. After the vehicles arrive, they are automatically checked and opened via smart locks, achieving a balance of data ledgers across the three ends.

7. The waste recycling method based on multidimensional vision and multidimensional weight estimation as described in claim 6, characterized in that, The cloud server dynamically allocates visual inspection threads, weighing platform operating frequency, and pneumatic nozzle operating sequence and energy consumption limits based on the digital twin sand table.

8. The waste recycling method based on multidimensional vision and multidimensional weight estimation as described in claim 2, characterized in that, The multispectral vision camera captures visible light and infrared spectral images, and the millimeter-wave radar captures three-dimensional contour point clouds. The data from both are fused at the feature level through a lightweight AI large model within the edge computing chip, thereby outputting the item type recognition result.

9. A non-volatile storage medium storing a computer program, which, when executed by a processor, implements the garbage collection method based on multidimensional vision and multidimensional weight estimation as described in any one of claims 1 to 8.

10. A waste recycling system based on multidimensional vision and multidimensional weight estimation, characterized in that, include: The identification module is used to process the results of the delivery item type identification. The multidimensional estimation and verification module is used to process the verification pricing results obtained from multidimensional data collection. The points settlement module is used to process user points settlement data from compliance judgment. The account withdrawal module is used to process user points settlement data into user account points withdrawal.

Images