Scalable ring sorters and methods based on visual pre-counting and weight verification
The scalable ring sorting device, which combines visual prediction and weighing verification, solves the problems of low space utilization and large counting deviation in traditional sorting systems, and achieves efficient and accurate sorting of multiple batches of materials and production continuity.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- FOSHAN HIGH TECH MACHINERY EQUIP
- Filing Date
- 2026-04-09
- Publication Date
- 2026-07-24
AI Technical Summary
Existing sorting systems suffer from low space utilization, large counting errors, poor equipment adaptability, and easy interruption of production processes when processing multiple batches of materials.
An scalable ring sorting device based on visual prediction and weighing verification is adopted. The multi-group sorting structure is arranged radially around the central feeder. Combined with the dual verification mechanism of AI visual prediction and weighing verification, the material is simultaneously sorted and verified in real time.
It optimizes space utilization, improves counting accuracy and production continuity, adapts to the production needs of different batches and scales, and reduces production downtime caused by equipment failure.
Smart Images
Figure CN122441663A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of industrial automation weighing and sorting technology, and more specifically, it relates to an scalable ring sorting device and method based on visual prediction and weighing verification. Background Technology
[0002] In the field of industrial automated production, material sorting and metering are key links in ensuring production efficiency and controlling product quality, and are widely used in food processing, pharmaceutical packaging, and hardware manufacturing industries. As the market demands for product precision and production efficiency continue to increase, traditional sorting equipment needs to meet the flexible processing requirements of multiple batches and small quantities.
[0003] Currently, most mainstream sorting systems in the industry adopt a single-channel linear layout or fixed unit structure. When dealing with materials of different specifications and large-scale batch production, they often suffer from insufficient scalability and low space utilization, making it difficult to adapt to the flexible requirements of modern production. At the same time, the accuracy of material metering directly affects product cost accounting and quality control. Existing metering methods mostly rely on single visual counting or weighing, which are easily affected by factors such as material shape and environmental interference, making it difficult to balance counting efficiency and metering accuracy, thus restricting the stability and reliability of the overall production process.
[0004] In practical applications, existing sorting systems with linear layouts require sequential conveying, metering, and sorting operations when processing multiple batches of materials. This process is cumbersome and occupies a large amount of space. Equipment with fixed unit structures cannot flexibly add or remove sorting channels according to production needs, resulting in poor adaptability. Furthermore, the single visual counting method is susceptible to the influence of material overlap, obstruction, and reflection, leading to counting deviations. The single weighing and metering method may result in inaccurate measurements due to differences in material weight and errors in weight sensors. Moreover, the lack of an effective data verification mechanism makes it impossible to detect and correct measurement errors in a timely manner. When a single sorting unit malfunctions, the lack of a redundant replacement mechanism can easily lead to the interruption of the entire production process, reducing production continuity. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides an scalable ring sorting device and method based on visual prediction and weighing verification. This addresses the issues in the prior art where traditional sorting devices occupy a large space, are prone to counting errors due to the use of a single counting method, and are susceptible to interruption of the entire production process when a single sorting unit malfunctions.
[0006] The objectives and effects of the scalable ring sorting device and method based on visual prediction and weighing verification of the present invention are achieved by the following specific technical means: An scalable ring sorting device based on visual prediction and weighing verification includes a control box. The top of the control box is equipped with a central feeder and a multi-group sorting structure. The multi-group sorting structure is arranged radially around the central feeder. The top of the central feeder is equipped with an upper receiving hopper. The multi-group sorting structure includes a video acquisition module for acquiring dynamic video data. A buffer hopper is provided on one side of the central sorting machine. A buffer chute is provided inside the buffer hopper. The video acquisition module is provided on the top of the buffer hopper facing the buffer chute.
[0007] According to a preferred embodiment, a central disc is provided below the upper receiving hopper, and multiple diversion plates are provided on the top of the central disc corresponding to multiple sets of sorting structures, with diversion grooves opened on the top of the diversion plates.
[0008] According to a preferred embodiment, the sorting structure further includes a rotary feeding device. The rotary feeding device is provided at the top of the diversion chute. The rotary feeding device includes a feeder. A first mounting plate is provided at the top of the diversion plate. The feeder is provided at the bottom of the first mounting plate corresponding to the diversion chute. A DC motor for driving the feeder is provided at the top of the first mounting plate. The DC motor and the feeder are connected by a belt structure.
[0009] According to a preferred embodiment, the diversion chute is inclined toward the buffer chute, and a vibrator is provided at the bottom of the diversion plate.
[0010] According to a preferred embodiment, a second mounting plate is provided on one side of the central feeder, and a weight sensor is provided on one side of the second mounting plate. The top of the weight sensor is connected to the buffer hopper through a first sensing bracket.
[0011] According to a preferred embodiment, a second sensing bracket is provided at the bottom of the weight sensor, and a weighing hopper is provided at the end of the second sensing bracket away from the weight sensor. A weighing groove is provided at the top of the weighing hopper corresponding to the discharge end of the buffer hopper.
[0012] According to a preferred embodiment, a discharge trough is provided on the top of the control box, and a material chute is provided on the top of the control box. The top of the material chute is provided with a material chute corresponding to the discharge end of multiple weighing hoppers, and the bottom end of the material chute passes through the discharge trough and is connected to the feeding hopper.
[0013] A scalable ring sorting method based on visual prediction and weighing verification, applied to the aforementioned scalable ring sorting device based on visual prediction and weighing verification, is characterized by comprising the following steps: S1: Based on the video acquisition module, acquire dynamic video data of the material sliding process, perform video frame parsing, multi-target tracking and counting fault tolerance based on the dynamic video data, and obtain AI vision prediction data; S2: Obtain material weighing data based on weight sensors, calculate the error between AI vision-predicted count data and material weighing data, and obtain counting result data; S3: Based on the counting results, the weighing hopper discharges material and the material chute is guided to obtain the qualified material after sorting.
[0014] According to a preferred embodiment, the step of acquiring dynamic video data of the material sliding process based on the video acquisition module, and performing video frame parsing, multi-target tracking, and counting error tolerance based on the dynamic video data to obtain AI visual prediction data includes: Using a video capture module with the top of the buffer hopper facing the buffer chute, dynamic video recording is performed on the material sliding into the buffer chute to obtain dynamic video data of the material sliding process; Dynamic video data of the material sliding process is decomposed into continuous frames. The material is identified by the small target detection layer of the improved YOLOv8 model. The Transformer attention module is used to enhance feature extraction. The material appearance features are extracted by combining MobileNetV3. The trajectory is predicted by Kalman filtering. When the material enters the counting area, counting is triggered. IoU+NMS dual filtering is used to process overlapping objects. A maximum of 15 frames of trajectory loss is allowed. The data is correlated across frames by feature similarity to obtain AI visual prediction data.
[0015] According to a preferred embodiment, the step of acquiring material weighing data based on a weight sensor, calculating the error between the AI visual prediction data and the material weighing data, and obtaining counting result data includes: The material slides into the buffer chute of the buffer hopper for temporary storage. The weight sensor on one side of the second mounting plate senses the overall weight of the buffer hopper through the first sensing bracket. After deducting the weight of the buffer hopper itself, the initial weight data of the material is obtained. Based on the initial weight data of the material, the buffer hopper opens the discharge end and lets all the temporarily stored material fall into the weighing trough at the top of the weighing hopper. The weight sensor senses the overall weight of the weighing hopper through the second sensing bracket, and after deducting the weight of the weighing hopper itself, the material weighing data is obtained. Based on AI visual prediction data and accurate material weighing data, the theoretical weight corresponding to the AI visual prediction is calculated according to the preset standard weight of a single material. The absolute error and relative error are calculated by comparing the AI visual prediction data with the accurate material weighing data. When the relative error is ≤5%, the visual prediction result is accepted. When 5% < relative error ≤15%, the material quantity is deduced from the weighing data. When the relative error >15%, an abnormal alarm is triggered, and the counting result data is obtained.
[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. By adopting a multi-group sorting structure in a radial, ring-shaped layout around the central distributor, the space utilization is optimized compared to traditional linear sorting devices. Multi-channel synchronous sorting operations can be achieved without a long linear conveying path, effectively reducing the overall production space occupied by the equipment and better meeting the needs of compact layouts in modern production workshops. At the same time, the ring-shaped layout, combined with the centrifugal force of the central disc, can evenly distribute materials to each sorting structure. Combined with the inclined design of the diversion chute and the synergistic effect of the vibrator and rotary feeding device, material accumulation and jamming during the conveying process can be avoided, ensuring smooth material conveying and further improving the continuity and overall efficiency of the sorting operation.
[0017] 2. Employing a dual-verification mechanism of AI visual counting and weighing verification, the system dynamically counts materials within the buffer chute using a video acquisition module, while simultaneously employs a weight sensor for precise weighing. These two sets of data mutually verify and correct each other, effectively overcoming the shortcomings of traditional single-counting methods. The visual counting module captures material dynamics in real time and performs intelligent identification and counting, while the weighing module infers the quantity from the material weight. This dual protection significantly reduces counting deviations caused by factors such as material overlap, occlusion, and individual weight differences, improving the accuracy and reliability of the counting results.
[0018] 3. It features a scalable multi-unit collaborative architecture, allowing for flexible adjustments to the number of sorting structures based on production needs, adapting to different batches and scales of production tasks, thus enhancing the equipment's versatility and adaptability. Simultaneously, the design of independent multi-unit operation, coupled with a fault-tolerant mechanism, ensures that if a single sorting structure malfunctions, the remaining sorting structures can continue operating normally without interrupting the entire production process. Furthermore, redundant units can be activated to replace the malfunctioning unit, effectively improving the system's fault tolerance and production continuity, and reducing production losses due to equipment failure. Attached Figure Description
[0019] Figure 1 This is a schematic diagram of the assembled structure of the present invention; Figure 2 This is a front view of the present invention; Figure 3 This is a top view of the present invention; Figure 4 This is a schematic diagram of the assembled sorting structure in this invention; Figure 5 This is a left view of the sorting structure in this invention; Figure 6 This is a flowchart of the steps of the scalable ring sorting method based on visual prediction and weighing verification of the present invention.
[0020] In the diagram, the correspondence between component names and drawing numbers is as follows: 11. Control chassis; 12. Central feeder; 13. Upper hopper; 14. Central disc; 15. Diverter plate; 16. Diverter chute; 17. Discharge chute; 18. Material sluice; 19. Material chute; 20. Feed hopper; 21. Video acquisition module; 22. Buffer hopper; 23. Buffer chute; 24. Feeder; 25. First mounting plate; 26. DC motor; 27. Belt structure; 28. Vibrator; 29. Second mounting plate; 30. Weight sensor; 31. First sensing bracket; 33. Second sensing bracket; 34. Weighing hopper; 35. Weighing trough. Detailed Implementation
[0021] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the technical solutions of the present invention, but should not be used to limit the scope of protection of the present invention.
[0022] Example:
[0023] As attached Figures 1 to 5 As shown: This invention provides an expandable circular sorting device based on visual prediction and weighing verification. It includes a control housing 11, which serves as the overall mounting base for the device, providing stable installation support and operational protection for all components. A central distributor 12 and multi-group sorting structures are arranged radially around the central distributor 12. This arrangement allows the central distributor 12 to evenly distribute the material to be sorted into each sorting structure, achieving multi-channel synchronous sorting. An upper receiving hopper 13 is located on top of the central distributor 12, serving to receive externally fed material and providing a guiding channel for the material entering the central distributor 12, preventing material feeding errors. In case of material spillage, each of the multi-group sorting structures includes a video acquisition module 21 for acquiring dynamic video data. The central feeder 12 is equipped with a buffer hopper 22 on one side facing each sorting structure. The buffer hopper 22 is equipped with a buffer chute 23, which is used to receive the material conveyed by the central feeder 12 and constrain the sliding trajectory of the material so that the material can slide smoothly along the preset path. The top of the buffer hopper 22 is equipped with a video acquisition module 21 facing the buffer chute 23. The video acquisition module 21 can continuously capture dynamic video of the material sliding in the buffer chute 23 and acquire dynamic video data during the material sliding process, providing basic data support for subsequent material visual counting operations.
[0024] A central disc 14 is provided below the upper hopper 13. The central disc 14 can rotate by itself and use centrifugal force to evenly disperse the material to be sorted from the upper hopper 13 to the corresponding positions of each sorting structure. The top of the central disc 14 is provided with multiple diversion plates 15 corresponding to the multiple sorting structures. The diversion plates 15 constrain the conveying direction of the material, so that the dispersed material can enter the conveying path of the corresponding sorting structure. The top of the diversion plate 15 is provided with a diversion chute 16, which provides a directional conveying channel for the material and guides the material to move towards the buffer hopper 22 of the corresponding sorting structure, so as to avoid the material from deviating from the path during the conveying process.
[0025] The sorting structure also includes a rotary feeding device. The rotary feeding device is located at the top of the diversion chute 16 and includes a feeder 24. A first mounting plate 25 is located at the top of the diversion plate 15, providing mounting support for each component of the rotary feeding device to ensure stable installation. The feeder 24 is located at the bottom of the first mounting plate 25 corresponding to the diversion chute 16, and a DC motor 26 for driving the feeder 24 is located at the top of the first mounting plate 25. The DC motor 26 and the feeder 24 are connected by a belt. The structure 27 is connected, and the DC motor 26 can output rotational power when running. The rotational power is transmitted to the feeder 24 through the belt structure 27, which drives the feeder 24 to rotate on the top of the diversion chute 16. During the rotation of the feeder 24, it can assist in pushing the material in the diversion chute 16 to avoid the accumulation and jamming of the material in the diversion chute 16. This allows the material to be continuously and stably conveyed along the diversion chute 16 towards the buffer hopper 22, ensuring the continuity of the material conveying process and meeting the continuous operation requirements of the sorting operation.
[0026] The diversion chute 16 is inclined toward the buffer chute 23. This inclination allows the material in the diversion chute 16 to slide continuously toward the buffer chute 23 under its own gravity, providing gravity assistance for the material to be conveyed from the diversion chute 16 to the buffer chute 23 and reducing the retention of material in the trough. A vibrator 28 is provided at the bottom of the diversion plate 15. After the vibrator 28 is activated, it can generate a continuous vibration force. This vibration force can be transmitted to the trough of the diversion chute 16 through the diversion plate 15, causing the material in the diversion chute 16 to vibrate synchronously and slightly, avoiding the stacking and jamming of materials in the diversion chute 16. At the same time, it can prevent materials from getting stuck in the gaps of the diversion chute 16. With the inclination of the diversion chute 16, it ensures that the material can be continuously and stably conveyed from the diversion chute 16 to the buffer chute 23, so that each sorting structure can obtain a uniform material supply, which is suitable for the operation rhythm of subsequent visual acquisition and weighing verification processes.
[0027] A second mounting plate 29 is provided on one side of the central feeder 12. This second mounting plate 29 provides stable support for the weight detection components, ensuring that each component remains in a fixed position during operation and does not shift or shake. A weight sensor 30 is installed on one side of the second mounting plate 29. The weight sensor 30 collects the weight signal transmitted by the corresponding component, providing basic data support for the material weight verification process. The top of the weight sensor 30 is connected to the buffer hopper 22 via a first sensing bracket 31. The first sensing bracket 31 can completely transmit the weight change of the buffer hopper 22 to the weight sensor 30. When material enters the buffer hopper 22 along the buffer chute 23 for temporary storage, the weight sensor 30 can collect preliminary weight data of the material in the buffer hopper 22 through the weight signal transmitted by the first sensing bracket 31. This data can be preliminarily cross-validated with the visual counting data acquired by the video acquisition module 21 to determine the buffer weight. To detect whether there is any abnormal retention of material in the hopper 22, a second sensing bracket 33 is provided at the bottom of the weight sensor 30. The second sensing bracket 33 can completely transmit the weight change of the weighing hopper 34 to the weight sensor 30. The weighing hopper 34 is provided at the end of the second sensing bracket 33 away from the weight sensor 30. The weighing hopper 34 is used to receive the material falling from the discharge end of the buffer hopper 22, providing a stable bearing space for the weight detection of the material. A weighing groove 35 is opened at the top of the weighing hopper 34 corresponding to the discharge end of the buffer hopper 22. The weighing groove 35 can receive and limit the material falling from the buffer hopper 22, preventing the material from scattering during the falling process and ensuring that all the material can fall into the bearing range of the weighing hopper 34. When all the material in the buffer hopper 22 has fallen into the weighing groove 35, the weight sensor 30 can collect the actual weight data of the material in the weighing hopper 34 through the weight signal transmitted by the second sensing bracket 33.
[0028] The top of the control box 11 is provided with a discharge channel 17, which provides through clearance space for the vertical conveying of materials and the installation of the material chute 18, avoiding interference from the control box 11 shell on the material discharge process. The top of the control box 11 is provided with a material chute 18, which provides a bearing base for the directional conveying of materials after sorting, ensuring that the materials output from each sorting structure can be discharged along a unified path. The top of the material chute 18 is provided with a material chute 19 corresponding to the discharge end of multiple weighing hoppers 34. After the weighing hopper 34 completes the weighing and verification process of the materials, the discharge end can be opened to release the materials inside, which will fall into the material chute 19 below. The material chute 19 constrains the sliding trajectory of the materials, preventing the materials from falling during the discharge process. In the event of material spillage or path deviation, multiple sets of corresponding material chutes 19 can simultaneously receive materials output from multiple weighing hoppers 34, enabling synchronous discharge of materials in multi-channel sorting operations. This adapts to the operational rhythm of multi-group sorting structures. The bottom end of the material chute 18 passes through the discharge channel 17 and connects to the discharge hopper 20. The discharge channel 17 ensures that the material chute 18 smoothly connects to the material conveying end at the top of the control box 11 and the discharge hopper 20 inside the control box 11. The discharge hopper 20 is used to receive the sorted materials conveyed from the material chute 19, uniformly collecting and guiding the materials gathered from multiple channels. This provides a stable discharge connection for subsequent material collection and storage processes, preventing material spillage at the discharge end and ensuring the orderly operation of the entire sorting process.
[0029] Please see as follows Figure 6 As shown, the present invention also provides a scalable ring sorting method based on visual prediction and weighing verification, applied to the aforementioned scalable ring sorting device based on visual prediction and weighing verification, comprising the following steps: S1: Based on the video acquisition module 21, acquire dynamic video data of the material sliding process, perform video frame parsing, multi-target tracking and counting fault tolerance based on the dynamic video data, and obtain AI visual prediction data; Specifically, the material to be sorted is fed into the upper receiving hopper 13, which provides a receiving and guiding space to prevent the material from scattering during the feeding process. The central disc 14 below the upper receiving hopper 13 starts to rotate, and the centrifugal force generated by the rotation evenly disperses the material falling from the upper receiving hopper 13 into the multi-group diversion plates 15. The diversion plates 15 constrain the conveying direction of the material, so that the dispersed material enters the conveying path of the corresponding sorting structure. The diversion chutes 16 at the top of the diversion plates 15 guide the material flow to the corresponding sorting structure in a radial layout, providing a directional channel for the material to be conveyed towards the buffer hopper 22 of the corresponding sorting structure, preventing the material from deviating from its path during the conveying process. The vibrator 28 at the bottom of the diversion plate 15 is activated. The vibration force generated by the operation of the vibrator 28 is transmitted through the diversion plate 15 to the trough of the diversion chutes 16, causing the material in the diversion chutes 16 to vibrate synchronously and slightly, preventing the material from piling up and getting stuck in the diversion chutes 16. In this situation, the DC motor 26 on the top of the first mounting plate 25 is started simultaneously. The DC motor 26 outputs rotational power, which is transmitted to the feeder 24 through the belt structure 27. The feeder 24 is driven to rotate on the top of the diversion chute 16. During the rotation, the feeder 24 assists in pushing the material in the diversion chute 16. With the diversion chute 16 tilted towards the buffer chute 23, the material is assisted to slide smoothly down the tilted diversion chute 16 and enter the buffer chute 23 in the buffer hopper 22. The buffer chute 23 constrains the sliding trajectory of the material, allowing the material to slide smoothly along the preset path, providing a stable material movement environment for subsequent video acquisition operations. The video acquisition module 21 on the top of the buffer hopper 22 facing the buffer chute 23 continuously records the dynamic video of the material sliding into the buffer chute 23, and obtains the dynamic video data of the material sliding process. This dynamic video data completely records the entire process of the material sliding in the buffer chute 23.
[0030] Based on the acquired dynamic video data of the material sliding process, the dynamic video is decomposed into continuous single-frame images. An improved YOLOv8 model's small target detection layer identifies the material within each frame, capturing features for smaller materials. A Transformer attention module enhances feature extraction capabilities, improving the completeness of material recognition. MobileNetV3 is used as the backbone network to extract the material's appearance features, providing a feature basis for subsequent multi-target tracking. Kalman filtering predicts the material's trajectory, triggering a counting action when the material enters a preset counting area. The IoU+NMS dual filtering method is used to process overlapping objects in the image, distinguish between mutually occluded material individuals, and avoid double counting or omission. The maximum trajectory loss time of 15 frames is allowed. When the trajectory is interrupted due to temporary occlusion of the material, the trajectory is correlated across frames through feature similarity to restore the complete material movement trajectory. Finally, the AI vision prediction data is obtained. This data can intuitively reflect the amount of material entering the buffer hopper 22, providing a prediction benchmark for the subsequent weighing and verification process. At the same time, it can be used for preliminary cross-validation with the weight data corresponding to the buffer hopper 22 to determine whether the material is stuck or abnormally falling.
[0031] S2: Based on the weight sensor 30, material weighing data is acquired, and the error is calculated between the AI vision prediction data and the material weighing data to obtain the counting result data; Specifically, the material slides into the buffer chute 23 of the buffer hopper 22 for temporary storage. The buffer chute 23 provides a stable storage space for the material, preventing continuous material slippage from interfering with the weight detection process. The second mounting plate 29 on one side of the central feeder 12 provides stable mounting support for the weight detection-related components, ensuring that the positions of each component are fixed during device operation and preventing displacement or shaking that could affect the detection data. The weight sensor 30 on one side of the second mounting plate 29 senses the overall weight of the buffer hopper 22 through the first sensing bracket 31. The first sensing bracket 31 can completely transmit the weight change of the buffer hopper 22 to the weight sensor 30, avoiding loss during weight signal transmission. After the weight sensor 30 collects the overall weight data of the buffer hopper 22, it subtracts the weight of the buffer hopper 22 itself to obtain preliminary weight data of the material. This data can be preliminarily cross-validated with the previously acquired AI visual prediction data to determine the weight of the material in the buffer hopper 22. The presence of abnormal material retention or leakage provides a preliminary reference for subsequent weighing verification. Based on the acquired preliminary material weight data, the buffer hopper 22 opens its discharge end, allowing all the temporarily stored material to fall into the weighing trough 35 at the top of the weighing hopper 34. The weighing trough 35 can receive and limit the material falling from the buffer hopper 22, preventing the material from scattering during the fall and ensuring that all material can enter the bearing range of the weighing hopper 34, thus ensuring the integrity of subsequent weight detection data. The weight sensor 30 senses the overall weight of the weighing hopper 34 through the second sensing bracket 33. The second sensing bracket 33 can completely transmit the weight change of the weighing hopper 34 to the weight sensor 30, ensuring the stability of the weight signal transmission. After the weight sensor 30 collects the overall weight data of the weighing hopper 34, it subtracts the weight of the weighing hopper 34 itself to obtain the material weighing data. This data is the actual weight detection result of the material, providing the core basis for the weight dimension for subsequent counting verification.Based on the previously acquired AI visual prediction data and material weighing data, and according to the preset standard weight of a single material, the theoretical weight corresponding to the AI visual prediction is calculated. This theoretical weight provides a reference benchmark for the weight dimension of the visual prediction results. Then, the absolute and relative errors are calculated by comparing the material weighing data. A graded error threshold is used to determine the counting results. When the relative error is ≤5%, it indicates a high degree of matching between the visual prediction results and the actual weight data, and the visual prediction results are accepted. When 5% < relative error ≤15%, it indicates a certain deviation in the visual prediction results, and the material quantity is deduced from the weighing data, correcting the result through weight dimension data. Deviations during visual recognition, with a relative error exceeding 15%, indicate a significant discrepancy between the predicted visual count and the actual weight data. This triggers an alarm, prompting operators to inspect the equipment's operational status and material conveying to prevent issues such as missed counts, jams, or spillage. The final count data combines the efficiency of visual recognition with the stability of weighing detection, reducing counting errors caused by material overlap, obstruction, and individual weight differences. This provides a reliable counting basis for subsequent quality control in material sorting. Furthermore, a tiered judgment method ensures accurate counting results while maintaining continuous sorting operations.
[0032] S3: Based on the counting results, the weighing hopper 34 discharges material and the material chute 18 guides it to obtain qualified material after sorting.
[0033] Specifically, if the counting results meet the preset requirements, it indicates that the counting results of this batch of materials meet the production specifications and are a qualified batch of materials. The weighing hopper 34 opens its discharge end, releasing the material it carries into the material chute 19 on top of the corresponding material chute 18 below. The material chute 18 provides a unified conveying and bearing foundation for the qualified materials output from the multi-group sorting structure. The material chute 19 on top of the material chute 18, corresponding to the discharge ends of multiple weighing hoppers 34, can receive and constrain the falling material, preventing the material from scattering or deviating from its path during release. At the same time, it can receive the material output synchronously from multiple weighing hoppers 34, realizing synchronous discharge of multi-channel sorting operations without interference between materials from different sorting units. The material slides down along the inclined material chute 19, passing through the discharge channel 17 opened on top of the control box 11. The discharge channel 17 is for the installation of the material chute 18 and the material... The vertical conveyor provides a through clearance space to prevent the housing of the control box 11 from interfering with the material's downward movement, ensuring that the material can be smoothly conveyed from the top of the control box 11 to the inside. After sliding down the material chute 19, the material finally collects into the discharge hopper 20. The discharge hopper 20 collects and guides the qualified materials from the multi-channel conveyor to the output, providing a stable discharge connection for subsequent material collection and storage processes, preventing the material from scattering at the discharge end. At the same time, it records the relevant data of the batch of materials and obtains the data of qualified materials after sorting. This data can be used for subsequent production statistics and quality traceability, providing complete closed-loop data support for production process control. If the counting result data does not meet the preset requirements, the weighing hopper 34 will not perform the discharge action of qualified materials, and will retain the abnormal materials of the corresponding batch, preventing materials that do not meet the counting requirements from entering the qualified material collection stage, and ensuring that the sorted output materials all meet the production specifications.
[0034] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.
Claims
1. A scalable ring sorting device based on visual prediction and weighing verification, characterized in that: It includes a control box (11), on the top of which is a central feeder (12) and a multi-group sorting structure. The multi-group sorting structure is arranged radially around the central feeder (12), and the top of the central feeder (12) is provided with an upper receiving hopper (13). The multi-group sorting structure includes a video acquisition module (21) for acquiring dynamic video data. A buffer hopper (22) is provided on one side of the central feeder (12). A buffer chute (23) is provided inside the buffer hopper (22). The video acquisition module (21) facing the buffer chute (23) is provided on the top of the buffer hopper (22).
2. The scalable ring sorting device based on visual prediction and weighing verification according to claim 1, characterized in that: A central disc (14) is provided below the upper receiving hopper (13). The top of the central disc (14) is provided with multiple diversion plates (15) corresponding to multiple sets of sorting structures. A diversion groove (16) is provided on the top of the diversion plate (15).
3. The scalable ring sorting device based on visual prediction and weighing verification according to claim 2, characterized in that: The sorting structure also includes a rotary feeding device. The rotary feeding device is provided on the top of the diversion chute (16). The rotary feeding device includes a feeder (24). A first mounting plate (25) is provided on the top of the diversion plate (15). The feeder (24) is provided on the bottom of the first mounting plate (25) corresponding to the diversion chute (16). A DC motor (26) for driving the feeder (24) is provided on the top of the first mounting plate (25). The DC motor (26) and the feeder (24) are connected by a belt structure (27).
4. The scalable ring sorting device based on visual prediction and weighing verification according to claim 2, characterized in that: The diversion chute (16) is inclined toward the buffer chute (23), and a vibrator (28) is provided at the bottom of the diversion plate (15).
5. The scalable ring sorting device based on visual prediction and weighing verification according to claim 1, characterized in that: A second mounting plate (29) is provided on one side of the central feeder (12), and a weight sensor (30) is provided on one side of the second mounting plate (29). The top of the weight sensor (30) is connected to the buffer hopper (22) through a first sensing bracket (31).
6. The scalable ring sorting device based on visual prediction and weighing verification according to claim 5, characterized in that: The weight sensor (30) is provided with a second sensing bracket (33) at the bottom end. The end of the second sensing bracket (33) away from the weight sensor (30) is provided with a weighing hopper (34). The top of the weighing hopper (34) is provided with a weighing groove (35) corresponding to the discharge end of the buffer hopper (22).
7. The scalable ring sorting device based on visual prediction and weighing verification according to claim 6, characterized in that: The top of the control box (11) is provided with a discharge channel (17) and a material chute (18) is provided on the top of the control box (11). The top of the material chute (18) is provided with a material chute (19) corresponding to the discharge end of multiple weighing hoppers (34). The bottom end of the material chute (18) passes through the discharge channel (17) and is connected to the feeding hopper (20).
8. A scalable ring sorting method based on visual prediction and weighing verification, applied to the scalable ring sorting device based on visual prediction and weighing verification as described in any one of claims 1 to 7, characterized in that, Includes the following steps: S1: Based on the video acquisition module (21), obtain dynamic video data of the material sliding process, perform video frame parsing, multi-target tracking and counting fault tolerance based on the dynamic video data, and obtain AI visual prediction data; S2: Obtain material weighing data based on weight sensor (30), calculate the error between AI vision prediction data and material weighing data, and obtain counting result data; S3: Based on the counting result data, the weighing hopper (34) discharges material and the material chute (18) guides the material to obtain the qualified material after sorting.
9. The scalable ring sorting device based on visual prediction and weighing verification according to claim 8, characterized in that, The dynamic video data of the material sliding process is acquired by the video acquisition module (21), and video frame parsing, multi-target tracking and counting fault tolerance are performed based on the dynamic video data to obtain AI visual prediction data, including: Using the video acquisition module (21) with the top of the buffer hopper (22) facing the buffer chute (23), the dynamic video of the material sliding into the buffer chute (23) is recorded to obtain dynamic video data of the material sliding process; Dynamic video data of the material sliding process is decomposed into continuous frames. The material is identified by the small target detection layer of the improved YOLOv8 model. The Transformer attention module is used to enhance feature extraction. The material appearance features are extracted by combining MobileNetV3. The trajectory is predicted by Kalman filtering. When the material enters the counting area, counting is triggered. IoU+NMS dual filtering is used to process overlapping objects. A maximum of 15 frames of trajectory loss is allowed. The data is correlated across frames by feature similarity to obtain AI visual prediction data.
10. The scalable ring sorting device based on visual prediction and weighing verification according to claim 8, characterized in that, The process involves acquiring material weighing data based on a weight sensor (30), calculating the error between the AI visual prediction data and the material weighing data, and obtaining the counting result data, including: The material slides into the buffer chute (23) of the buffer hopper (22) for temporary storage. The weight sensor (30) on one side of the second mounting plate (29) senses the overall weight of the buffer hopper (22) through the first sensing bracket (31). After deducting the weight of the buffer hopper (22) itself, the material's initial weight data is obtained. Based on the preliminary weight data of the material, the buffer hopper (22) opens the discharge end and all the temporarily stored material falls into the weighing groove (35) at the top of the weighing hopper (34). The weight sensor (30) senses the overall weight of the weighing hopper (34) through the second sensing bracket (33), and after deducting the weight of the weighing hopper (34) itself, the material weighing data is obtained. Based on AI visual prediction data and accurate material weighing data, the theoretical weight corresponding to the AI visual prediction is calculated according to the preset standard weight of a single material. The absolute error and relative error are calculated by comparing the AI visual prediction data with the accurate material weighing data. When the relative error is ≤5%, the visual prediction result is accepted. When 5% < relative error ≤15%, the material quantity is deduced from the weighing data. When the relative error >15%, an abnormal alarm is triggered, and the counting result data is obtained.