Material box sorting data processing method and system based on visual weight fusion

Abstract

The application discloses a material box sorting data processing method and system based on visual weight fusion, and relates to the technical field of data processing. The method comprises the following steps: S1, collecting image data, weighing data and corresponding position and time information of the material box, and performing unified time base alignment and preprocessing; S2, analyzing the motion trajectory relationship of the material box on the disc, correlating and matching the visual data and the weighing data, and performing multi-source data binding; S3, analyzing the appearance integrity and the weight consistency, performing abnormality determination on the overall state of the material box, and performing material box quality detection; and S4, generating a sorting decision, and controlling the material box to perform classification and sorting at a downstream station. The problems of incorrect binding of multi-source information of the material box caused by the time sequence misalignment of the visual data and the weighing data in the disc sorting process are solved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for processing box sorting data based on visual weight fusion. Background Technology

[0002] With the continuous development of automated manufacturing and intelligent production lines, the demand for high-speed sorting of box-type workpieces in scenarios such as electronic assembly, pharmaceutical packaging, and precision machining is constantly increasing. During the production process, it is usually necessary to simultaneously inspect the appearance and loading weight of the boxes to ensure product consistency and stable production cycle. A multi-station collaborative approach based on visual inspection and weighing detection has become an important technical means for realizing box quality judgment and sorting control, and it has wide applications in multi-source data fusion processing and real-time decision control.

[0003] For example, Chinese patent CN121030318A discloses a logistics sorting method, system, equipment, and storage medium, belonging to the field of data processing technology. The method includes: acquiring the appearance inspection data, weight inspection data, and spectral inspection data corresponding to each of M fruits to be sorted in the current inspection batch; performing dimensionality reduction processing on the spectral inspection data based on a pre-constructed load matrix to obtain multiple sets of principal component features in the spectral inspection data; and sorting the M fruits to be sorted based on the multiple sets of principal component features in the appearance inspection data, weight inspection data, and spectral inspection data. The logistics sorting method, system, equipment, and storage medium provided in this application can improve the accuracy of fruit sorting.

[0004] For example, Chinese patent CN119089146B discloses a machine vision-based system for sorting and intelligent iron removal of anchor bolts in ore chutes, involving the field of image data processing. It includes: an anchor bolt sensing module for determining the presence of anchor bolts based on images, magnetic field information, sound information, and vibration information from the ore chutes; an ore chutes control module for stopping the ore chutes after determining the presence of anchor bolts; and an intelligent iron removal module for determining anchor bolt information based on images, magnetic field information, sound information, and vibration information from the ore chutes, acquiring images of the target area, and determining the presence of anchor bolts based on the target area images. After determining the presence of anchor bolts, it determines anchor bolt sorting parameters based on the target area images, and a collaborative robot performs anchor bolt sorting according to these parameters. The ore chutes control module also controls the ore chutes to continue operating after determining the absence of anchor bolts or after the collaborative robot has completed anchor bolt sorting, thus achieving the advantage of automated removal of anchor bolts mixed in with the ore.

[0005] However, in the automated disc sorting process, the boxes move continuously between workstations, and visual and weighing data acquisition are triggered independently by different devices. This is affected by fluctuations in mechanical cycle time, communication delays, and differences in control response, causing a time-series offset between the image and weight data of the same box. When the MCU or host computer performs data fusion, it may incorrectly bind data from adjacent boxes, leading to distorted quality judgment results and instances of misjudgment or omission. This is particularly prominent in high-speed production scenarios, severely impacting sorting accuracy and system reliability.

[0006] Therefore, in order to address the above problems, there is an urgent need for a data processing method and system for box sorting based on visual weight fusion. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method and system for processing material box sorting data based on visual weight fusion, which solves the problem of incorrect binding of multi-source information of material boxes caused by the time sequence misalignment of visual data and weighing data during the disc sorting process.

[0008] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a data processing method for box sorting based on visual-weight fusion, comprising the following steps: S1, real-time acquisition of visual data, weighing data, and corresponding position and time information of boxes in the disc line; unified time-base alignment and preprocessing of multi-source data; extraction of box appearance area features and stable weight values ​​to form a standardized box feature set; the stable weight value is used to reflect the actual loading mass of the box; S2, analysis of the movement trajectory relationship of the box on the disc using the standardized box feature set to obtain temporal and spatial position features; and combining the temporal and spatial position features with visual-weight fusion. S3. Visual data and weighing data are correlated and matched. Based on the correlation and matching results, multi-source data are bound to obtain fused data. Based on the fused data, the appearance integrity is analyzed using the effective target area area and the defect area area. The weight consistency is analyzed using the stable weight value. The overall state of the material box is judged for anomalies by combining appearance integrity and weight consistency. The quality of the material box is judged based on the anomaly judgment results. The effective target area area is used to reflect the effective occupancy of the material box visually. S4. Based on the material box quality judgment results, sorting decisions are generated to control the material boxes to perform classification and sorting at downstream workstations and to provide feedback on the sorting results.

[0009] Furthermore, the specific process of real-time acquisition of visual data, weighing data, and corresponding position and time information of the boxes in the disc line is as follows: As the boxes rotate with the disc and pass through each inspection station in sequence, real-time acquisition of box sorting data is performed, including: using an industrial camera at the visual inspection station to acquire images of the passing boxes to obtain visual data, including the image frame of the corresponding box and the visual acquisition timestamp; using a weighing sensor at the weighing station to continuously weigh the passing boxes to obtain weighing data, including the weighing value and the weighing acquisition timestamp; and simultaneously, the encoder continuously samples during the disc rotation process to obtain the disc corner position corresponding to each sampling moment.

[0010] Furthermore, the specific process of performing unified time-base alignment and preprocessing on multi-source data to extract the appearance area features and stable weight values ​​of the material box, forming a standardized material box feature set, is as follows: The visual acquisition timestamp and weighing acquisition timestamp are uniformly mapped to the standard clock reference of the production line for time alignment; Gaussian filtering is used to denoise the image frames, and histogram equalization is performed on the denoised image frames for gray-level equalization; based on threshold segmentation and connected component analysis methods, the region where the material box is located in the image frame is located to obtain the pixel set of the target area of ​​the material box, and the number of pixels in the set is counted to calculate the effective target area area of ​​the material box; within the target area of ​​the material box, based on gray-level difference and edge feature segmentation methods, abnormal areas are identified, and the pixel set of the abnormal areas is determined as the defect area pixel set, and the number of pixels in the set is counted to calculate the defect area area; the weighing values ​​of adjacent sampling points are calculated. The difference is used to obtain the weight change. When the weight change of two consecutive sampling points is greater than the judgment threshold, the corresponding sampling point is marked as a candidate position for the weighing start point. When the weight change of two consecutive sampling points is less than the negative of the judgment threshold, the corresponding sampling point is marked as a candidate position for the weighing end point. Between the candidate positions for the weighing start point and the candidate positions for the weighing end point, the sampling segment with a weight value continuously higher than the empty load threshold and the number of consecutive sampling points meets the minimum sampling point threshold is extracted as the weighing data segment corresponding to the material box. Within the weighing data segment corresponding to each material box, median filtering is performed on the weight value sequence, and based on the robust anomaly detection method of median and median absolute deviation, abnormal weight values ​​are identified and removed. The median of the remaining weight values ​​is taken as the stable weight value of the material box. Normalization processing is performed on the material box sorting data, and a material box sorting database is established to store the original and preprocessed material box sorting data.

[0011] Furthermore, the specific process of analyzing the motion trajectory relationship of the material box on the disc using the standardized material box feature set to obtain the temporal and spatial position features is as follows: For each visual data, the visual acquisition timestamp and the disc corner position at the corresponding sampling time are read, and the difference between the current visual acquisition timestamp and each weighing acquisition timestamp is calculated. Weighing data with differences less than the time difference threshold are selected as candidate weighing data sets. The weighing acquisition timestamp and the disc corner position at the corresponding sampling time for each weighing data in the candidate weighing data set are read. Based on the known structural conditions of the disc equipment, the fixed mechanical installation angle difference between the visual inspection station and the weighing station is obtained. Based on the material boxes that have been bound and sorted in the past, the time difference sequence between the weighing sampling timestamp and the visual sampling timestamp is statistically analyzed. The median value of the time difference sequence is taken as the typical migration time, and the median absolute deviation of the time difference sequence is calculated as the time fluctuation value. At the same time, the difference sequence of the disc corner positions from the historical visual inspection station to the weighing station is statistically analyzed, and the median absolute deviation is calculated to obtain the angular displacement fluctuation value.

[0012] Furthermore, the specific process of associating and matching visual data and weighing data by combining temporal and spatial features is as follows: For each weighing data in the candidate weighing data set, the actual time difference is obtained by subtracting the current visual acquisition timestamp from the corresponding weighing acquisition timestamp; the time deviation value is obtained by subtracting the typical migration time from the actual time difference; the normalized time deviation is obtained by dividing the time deviation value by the sum of the time fluctuation value and a preset positive number; the difference between the disk angle position corresponding to the weighing data sampling time and the disk angle position corresponding to the current visual data sampling time is calculated to obtain the actual angular displacement difference; the fixed mechanical installation angle difference is subtracted from the actual angular displacement difference to obtain the angular displacement deviation value; the normalized angular displacement deviation is obtained by dividing the angular displacement deviation value by the sum of the angular displacement fluctuation value and a preset positive number; the square root of the sum of the squares of the normalized time deviation and the normalized angular displacement deviation is obtained to obtain the association residual value between the visual data and the candidate weighing data.

[0013] Furthermore, the specific process of binding multi-source data based on the association matching results to obtain fused data is as follows: In the candidate weighing data set corresponding to the current visual data, the corresponding association residual value is calculated for each weighing data, and the weighing data with the smallest association residual value is selected as the optimal matching result; the stable weight value, weighing acquisition timestamp, and disk corner position corresponding to the optimally matched weighing data are bound to the current visual data to generate a hopper-level visual weight alignment data record, thus obtaining the fused data of the hopper.

[0014] Furthermore, based on the fused data, the process of analyzing appearance integrity using the effective target area area and defect area area, analyzing weight consistency using stable weight values, and combining appearance integrity and weight consistency to determine the overall state of the material box for anomaly judgment is as follows: For each material box, the corresponding fused data is read, and according to the product category to which the material box belongs, the reference effective area area and reference weight value of the corresponding category are read; the stable weight value of the current material box is divided by the reference weight value of the corresponding category to obtain the weight ratio, and a preset positive number is added to the weight ratio to obtain the weight item; the effective target area area of ​​the current material box is divided by the reference effective area area of ​​the corresponding category to obtain the area ratio; the defect area area of ​​the current material box is divided by the sum of the effective target area area and the preset positive number to obtain the defect percentage value, and one is subtracted from the defect percentage value to obtain the non-defect percentage value; the area ratio is multiplied by the non-defect percentage value to obtain the visual comprehensive percentage value; the square root of the visual comprehensive percentage value is taken and a preset positive number is added to obtain the visual item; the weight item is divided by the visual item, and the natural logarithm is taken, and the absolute value of the natural logarithm result is taken to obtain the visual weight consistency deviation value of the material box.

[0015] Furthermore, the specific process for judging the quality of the material box based on the anomaly judgment result is as follows: the visual weight consistency deviation value of the material box is compared with the deviation threshold to judge the quality: when the visual weight consistency deviation value is greater than the deviation threshold, the material box is judged as an abnormal material box; when the visual weight consistency deviation value is less than or equal to the deviation threshold, the material box is judged as a normal material box; at the same time, the defect area of ​​the corresponding material box is read. If the defect area is greater than the defect area threshold, the material box is judged to have an appearance abnormality and the material box is directly marked as an abnormal material box.

[0016] Furthermore, the specific process of generating sorting decisions based on the quality judgment results of the boxes, controlling the boxes to perform classification and sorting at downstream workstations, and providing feedback on the sorting results is as follows: After the box completes the quality judgment, the box is transferred from the disc-bearing workstation to the downstream conveying path. When the box reaches the sorting workstation, a sorting control command is generated: When the box is judged to be a normal box, the sorting execution mechanism is controlled to transport the box to the target grid in the preset multi-station sorting grid; when the box is judged to be an abnormal box, the sorting execution mechanism is controlled to transport the box to the waste trough; after sorting is completed, the sorting execution time, sorting execution status, and sorting destination of the current box are recorded and associated with the fusion data of the corresponding box to form a box sorting result record. Based on the fusion data of the sorted boxes, the reference weight value, reference effective area area, typical migration time, time fluctuation value, and angular displacement fluctuation value are updated.

[0017] The second aspect of this invention provides a bin sorting data processing system based on visual-weight fusion, comprising: a sorting data acquisition and processing module, used to acquire visual data, weighing data, and corresponding position and time information of bins in a circular conveyor belt in real time; perform unified time-base alignment and preprocessing on multi-source data; extract bin appearance area features and stable weight values ​​to form a standardized bin feature set; the stable weight value is used to reflect the actual loading mass of the bin; and a bin trajectory alignment module, used to analyze the motion trajectory relationship of the bins on the circular conveyor belt using the standardized bin feature set, obtain time features and spatial position features, and combine the time features and spatial position features to process the visual data and weighing data. The system includes a line association matching module, which binds multi-source data based on the association matching results to obtain fused data; a visual weight fusion evaluation module, which analyzes appearance integrity using the effective target area area and defect area area based on the fused data, analyzes weight consistency using stable weight values, and combines appearance integrity and weight consistency to determine the overall state of the box for anomalies. Based on the anomaly determination results, the system determines the quality of the box. The effective target area area is used to reflect the effective occupancy of the box visually; and a sorting decision feedback module, which generates sorting decisions based on the box quality determination results, controls the boxes to be sorted and classified at downstream workstations, and provides feedback on the sorting results.

[0018] Beneficial effects The present invention has the following beneficial effects: (1) This invention, by constructing a joint association matching mechanism based on time features and spatial location features, realizes accurate correspondence between visual data and weighing data under cross-workstation conditions, effectively avoids the problem of mismatch of material box data caused by the time sequence offset of acquisition, and improves the accuracy of multi-source data fusion.

[0019] (2) This invention improves the stable characterization of the appearance and weight of the box by performing segmented recognition and robust processing on the weighing data and combining it with the extraction of appearance area features. It effectively suppresses the interference of mechanical vibration and sampling noise on the detection results and improves the reliability of feature extraction.

[0020] (3) This invention, by constructing a fusion evaluation mechanism based on appearance integrity and weight consistency, realizes a comprehensive judgment of the overall state of the material box, and can simultaneously identify multiple types of problems such as missing parts, foreign objects and assembly abnormalities, thereby improving the comprehensiveness and accuracy of quality judgment.

[0021] (4) This invention combines the quality judgment result with sorting control and historical data update to form a closed-loop mechanism for sorting execution and data feedback. It can dynamically adjust the reference features and judgment range according to the running data, thereby improving the stability and adaptability under complex working conditions.

[0022] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0023] Figure 1 This is a flowchart of a data processing method for box sorting based on visual weight fusion. Figure 2 This is a structural diagram of a box sorting data processing system based on visual weight fusion. Figure 3 This is a schematic diagram showing the distribution of visual weight consistency deviations in the material box. Figure 4 This is a schematic diagram of the actual operation of box sorting based on visual weight fusion. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. As those skilled in the art will understand, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figures 1-4 This invention provides a technical solution: a method for processing box sorting data based on visual weight fusion, such as... Figure 1 As shown, the process includes the following steps: S1, real-time acquisition of visual data, weighing data, and corresponding position and time information of the boxes in the disc line; unified time base alignment and preprocessing of multi-source data; extraction of appearance area features and stable weight values ​​of the boxes to form a standardized box feature set; the stable weight value is used to reflect the actual loading mass of the boxes; S2, analysis of the movement trajectory relationship of the boxes on the disc using the standardized box feature set to obtain time features and spatial position features; association matching of visual data and weighing data based on time features and spatial position features; binding of multi-source data based on the association matching results to obtain fused data; S3, based on the fused data, analysis of appearance integrity using the effective target area area and defect area area; analysis of weight consistency using the stable weight value; anomaly judgment of the overall state of the boxes based on appearance integrity and weight consistency; quality judgment of the boxes based on the anomaly judgment results; the effective target area area is used to reflect the effective occupancy of the boxes visually; S4, generation of sorting decisions based on the box quality judgment results; control of the boxes to perform classification and sorting at downstream workstations; and feedback of sorting results.

[0026] Specifically, the real-time acquisition process of visual data, weighing data, and corresponding position and time information of the boxes in the disc conveyor is as follows: As the boxes rotate with the disc and pass through each inspection station sequentially, real-time box sorting data is acquired. The disc is a continuously rotating turntable structure with multiple support stations. Each support station is used to fix a single box, and the box maintains its position relative to the disc during rotation, thus achieving stable transfer of the box between inspection stations. The disc rotation speed is adjustable and stable continuous rotation to ensure that the movement of each box between different stations has a predictable time and angular relationship. This includes: using an industrial camera on the visual inspection station to acquire images of the passing boxes. The industrial camera is fixedly installed at a preset position on the disc path and connected to a trigger control unit. When a box is detected entering the visual acquisition area, it triggers an optical sensor... The electronic sensor signal triggers the image capture, ensuring that each material box corresponds to only one valid image acquisition, obtaining visual data, including the image frame of the corresponding material box and the visual acquisition timestamp. The acquisition timestamp is generated by a unified clock source, preferably from the PLC master station clock, to ensure the consistency of the time stamp. The weighing station uses a weighing sensor at another fixed position on the rotating path of the disc to continuously weigh the passing material boxes. The weighing sensor preferably forms force contact with the material box through a support structure, continuously outputting a weighing signal throughout the entire time period when the material box passes through the weighing area, thereby forming a weighing data sequence that varies with time, obtaining weighing data, including the weighing value and the weighing acquisition timestamp. The weighing data sampling frequency is preferably 50Hz to 500Hz to ensure that a sufficiently dense sampling point can be obtained when the material box passes through the weighing area for subsequent stable section extraction. Simultaneously, the encoder continuously samples during the disk's rotation. The encoder is connected to the disk drive shaft and outputs pulse signals that correspond one-to-one with the disk's rotation angle in real time. These signals are then converted into disk angular positions through counting and calibration, thereby establishing a correspondence between time and spatial position. The disk angular position at each sampling moment is obtained, which characterizes the spatial position of the material box at each sampling moment and serves as the basic parameter for subsequent spatiotemporal correlation matching of visual data and weighing data.

[0027] In this implementation scheme, visual data, weighing data, and corresponding angular position information of the material boxes are acquired simultaneously during the disc transfer process, enabling accurate acquisition and representation of multi-source data under a unified spatiotemporal reference. This makes the status of the material boxes at different inspection stations correlated and traceable. By combining trigger control with continuous sampling, the integrity and stability of data acquisition are ensured, thereby providing a reliable data foundation for the accurate matching of visual and weight data and quality judgment, and improving the accuracy and stability of the sorting process.

[0028] Specifically, the process of unifying time base alignment and preprocessing of multi-source data to extract the appearance features and stable weight values ​​of the material box and form a standardized material box feature set is as follows: The visual acquisition timestamp and weighing acquisition timestamp are uniformly mapped to the production line standard clock reference for time alignment. The production line standard clock reference is uniformly provided by the PLC. Each acquisition device includes a local timestamp when uploading data and is corrected through a time synchronization protocol, thus ensuring the comparability of data from different devices under the same time reference system. Gaussian filtering is used to denoise the image frames. Gaussian filtering suppresses random noise interference by performing convolution operations on the image with a set standard deviation parameter. Histogram equalization is used to perform grayscale equalization on the denoised image frames to enhance image contrast and improve the stability of subsequent segmentation. Threshold segmentation and connected component analysis are used to locate the area where the material box is located in the image frame. Threshold segmentation is used to initially separate the foreground and background, and connected component analysis is used to improve the localization of the foreground and background. The largest continuous pixel region is selected as the target region of the material box, resulting in a pixel set of the target region. The effective target region area of ​​the material box is calculated by counting the number of pixels in the set. Within the target region of the material box, a segmentation method based on grayscale difference and edge features is used. Grayscale difference is used to detect regions with sudden brightness changes, and edge features are extracted through operators to enhance defect recognition capabilities, identifying abnormal regions. The pixel set of abnormal regions is determined as the pixel set of defect regions, and the area of ​​the defect region is calculated by counting the number of pixels in the set. The effective target region area and the defect region area are represented by the number of pixels. In this embodiment, since the installation position, imaging parameters, and field of view of the image acquisition device remain unchanged, the spatial scale corresponding to each pixel is consistent. Therefore, the number of pixels is used to represent the area. In the subsequent calculation process, the area ratio is used to process the pixel size so that they cancel each other out in the calculation. Thus, the dimensionless expression of features can be achieved without actual physical area conversion.The difference in weighing values ​​between adjacent sampling points is calculated to obtain the weighing change. Adjacent sampling points are acquired according to a fixed sampling period of the weight signal to reflect the instantaneous trend of the weight signal. When the weighing changes of two consecutive sampling points are both greater than the judgment threshold, the corresponding sampling point is marked as a candidate position for the weighing start point; when the weighing changes of two consecutive sampling points are both less than the negative of the judgment threshold, the corresponding sampling point is marked as a candidate position for the weighing end point. The judgment threshold is obtained based on the historical sampling data of the weighing sensor under stable operating conditions. By analyzing the weighing change sequences during the passage of the material box with and without a material box, the statistical characteristics of the change distribution are extracted. The median value and median absolute deviation of the change are preferred to determine the threshold value, so that the judgment threshold can reflect the entry and exit of the material box from the weighing area. The typical variation level at time; between the candidate positions of the weighing start and the candidate positions of the weighing end, the sampling segment with a continuous weighing value higher than the empty load threshold and the number of continuous sampling points meeting the minimum sampling point threshold is extracted as the weighing data segment corresponding to the material box; the empty load threshold is determined based on the historical median value and median absolute deviation of the weighing sensor in the empty box state, and is used to distinguish between effective weighing and empty load noise; by setting the judgment threshold and the joint judgment of continuous sampling points, the false triggering problem caused by mechanical vibration, sensor noise and instantaneous disturbance can be effectively suppressed, and the stability and accuracy of the identification of the start position and the end position can be improved; at the same time, by constraining the minimum sampling point threshold, it is ensured that the extracted data segment has a sufficient length, thereby avoiding incorrect segmentation due to short-term fluctuations and improving the reliability of the weighing data segment extraction. Within each bin's corresponding weighing data segment, median filtering is performed on the weighing value sequence to eliminate the impact of sudden pulse interference on the data. A robust anomaly detection method based on the median and median absolute deviation is used to identify and remove abnormal weighing values. The median absolute deviation measures the stability of the data distribution by the degree to which statistical data deviates from the median, thus ensuring reliable identification of outliers. The median of the remaining weighing values ​​is taken as the stable weight value of the bin. Normalization processing is performed on the bin sorting data. Data from different sources are transformed into dimensionless values ​​according to their range. Area features and weighing values ​​are linearly normalized according to their maximum and minimum value ranges, where the maximum and minimum values ​​are statistically obtained from historical sample data of the corresponding bin product category. The disc corner positions are periodically normalized according to the angle range corresponding to one circumference of the disc, facilitating subsequent fusion calculations and consistency assessments. A bin sorting database is established to store raw and pre-processed bin sorting data according to bin number to support subsequent data traceability and model updates.

[0029] In this implementation scheme, visual and weighing data are preprocessed with unified time-base alignment and standardization. Combined with image denoising and region segmentation, stable extraction of the appearance features of the material box is achieved. Reliable and stable weight features are extracted from continuous weighing signals through threshold determination and segment recognition methods based on statistical features, thereby effectively suppressing the effects of mechanical vibration, sensor noise, and instantaneous disturbances. At the same time, normalization processing is used to unify the numerical scale of different features, and a structured data storage system is established to ensure consistency and comparability of multi-source data. This provides an accurate and stable data foundation for subsequent visual and weight fusion determination, improving the accuracy and reliability of the sorting process.

[0030] Specifically, the process of analyzing the motion trajectory relationship of the material boxes on the disk using a standardized material box feature set to obtain temporal and spatial features is as follows: For each visual data point, the visual acquisition timestamp and the disk corner position at the corresponding sampling time are read, and the difference between the current visual acquisition timestamp and each weighing acquisition timestamp is calculated. Weighing data with differences less than a time difference threshold are selected as candidate weighing data sets. During the candidate set selection, a spatial continuity constraint can be optionally introduced, retaining only data that satisfies a monotonic adjacency relationship with the current visual data in the disk rotation direction. The weighing acquisition timestamp and the disk corner position at the corresponding sampling time for each weighing data point in the candidate weighing data set are read. By setting a time difference threshold for preliminary screening of weighing data, data from distant time periods that are unlikely to belong to the same material box can be excluded, thus narrowing the matching range, reducing the complexity of subsequent correlation calculations, avoiding cross-material box mismatches, and improving the accuracy and stability of binding visual data and weighing data. Based on the known structural conditions of the disc-shaped equipment, the fixed mechanical installation angle difference between the vision inspection station and the weighing station is obtained. This fixed mechanical installation angle difference is the relative angular position difference between the vision inspection station and the weighing station in the circumferential direction of the disc. It is obtained through the calibration process after equipment installation. Specifically, this is achieved by reading the angular position data of the fixed encoders corresponding to the vision inspection station and the weighing station, calculating the difference, and writing the fixed mechanical installation angle difference into the configuration as a calibration parameter. Based on historically bound and sorted boxes, the time difference sequence between the weighing sampling timestamp and the vision sampling timestamp is statistically analyzed. The median value of the time difference sequence is taken as the typical migration time, and the median absolute deviation of the time difference sequence is calculated as the time fluctuation value. The time difference sequence originates from historically matched box samples to ensure that the typical migration time reflects the stable time characteristics of the box moving from the vision station to the weighing station under the current operating cycle. Simultaneously, the difference sequence of disk angular positions from the historical visual inspection station to the weighing station is statistically analyzed, and the median absolute deviation is calculated to obtain the angular displacement fluctuation value. The difference sequence of disk angular positions reflects the angular position deviation caused by mechanical micro-vibration, speed fluctuation, and encoder quantization error during actual operation. Statistical analysis using the median absolute deviation can characterize the normal fluctuation range of the deviation, thus providing a robust spatial constraint basis for subsequent correlation matching. Among them, the temporal characteristics include typical migration time and temporal fluctuation value, and the spatial position characteristics include fixed mechanical installation angular difference and angular displacement fluctuation value.

[0031] In this implementation plan, pre-screening of weighing data based on time difference thresholds effectively narrows the candidate matching range, reduces computational complexity, and avoids mismatches across material boxes. By combining the fixed mechanical installation angle difference obtained through calibration, a stable spatial constraint relationship between workstations is established. Furthermore, by utilizing the typical migration time and its fluctuation range and angular displacement fluctuation characteristics obtained from historical sample statistics, a stable motion law model of the material boxes in the time and space dimensions is achieved. This improves the accuracy and robustness of the association matching between visual data and weighing data under complex working conditions, and enhances the overall reliability of the system's sorting judgment.

[0032] Specifically, the process of linking and matching visual data and weighing data by combining temporal and spatial characteristics is as follows: For each weighing data in the candidate weighing data set, the actual time difference is obtained by subtracting the current visual acquisition timestamp from the corresponding weighing acquisition timestamp. The actual time difference is used to characterize the original offset of the candidate weighing data relative to the current visual data on the time axis. The time deviation value is obtained by subtracting the typical migration time from the actual time difference. The time deviation value is used to characterize the degree of deviation between the actual running time and the theoretical running time of the current material box, thereby reflecting the abnormal operation of the material box under the current cycle conditions. The time deviation value is divided by the sum of the time fluctuation value and the preset positive number to obtain the normalized time deviation. The time fluctuation value is used to characterize the statistical fluctuation range of the time deviation under normal operating conditions. By normalizing the time deviation, the time differences under different operating cycle conditions can be compared on a uniform scale, thereby avoiding the influence of production cycle changes on the matching results. The difference between the disk angle position at the time of weighing data sampling and the disk angle position at the time of current visual data sampling is calculated to obtain the actual angular displacement difference. This actual angular displacement difference characterizes the original spatial offset between the candidate weighing data and the visual data. Subtracting the fixed mechanical installation angle difference from the actual angular displacement difference yields the angular displacement deviation value. This value characterizes the degree of deviation between the actual spatial position of the current material box and the theoretical workstation position, reflecting the positional offset caused by mechanical errors and operational disturbances. Dividing the angular displacement deviation value by the sum of the angular displacement fluctuation value and a preset positive number yields the normalized angular displacement deviation. This fluctuation value characterizes the statistical fluctuation range of spatial position deviation under normal operating conditions. Normalizing the angular displacement deviation allows spatial deviations caused by mechanical vibration, encoder errors, etc., to be characterized on a uniform scale, thereby enhancing the robustness of spatial matching. The square root of the sum of the squares of the normalized time deviation and the normalized angular displacement deviation is then calculated to obtain the correlation residual value between the visual data and the candidate weighing data. The sum of squares operation is used to comprehensively integrate temporal and spatial deviations, avoiding the dominance of single-dimensional deviations in the matching results. The square root is used to restore the comprehensive deviation measure consistent with the original scale, making the results more interpretable. The correlation residual value is used to comprehensively characterize the overall deviation of candidate weighing data from the current visual data in both the temporal and spatial dimensions. By fusing temporal and angular displacement deviations at a unified scale, a multi-feature consistency matching criterion is constructed, thus transforming the originally scattered temporal and spatial matching processes into a single quantitative evaluation index. This method can preferentially select the weighing data most consistent with the current visual data even in the presence of cycle time fluctuations, equipment errors, and signal noise, achieving accurate binding of multi-source data for the same material box, effectively reducing the probability of mismatch and improving the stability and reliability of the sorting system under complex working conditions.

[0033] The specific formula for the associated residual value is as follows: ; In the formula, The associated residual value is used to jointly evaluate the temporal and spatial consistency between visual sampling data and candidate weighing data. The associated residual value is obtained by normalizing and fusing the time deviation and angular displacement deviation. The smaller the associated residual value, the more likely the two data are to originate from the same material box, thus enabling accurate matching of multi-source data across workstations. This represents the timestamp of the weighing data collection, used to reflect the moment the weighing occurred; This represents the current visual acquisition timestamp, used to reflect the moment the image was acquired; This represents the typical migration time, serving as a benchmark for characterizing the time interval of the cartridge under normal operation. It represents the time fluctuation value, used to measure the normal scale of time deviation and to normalize the time term; This indicates the disk corner position corresponding to the sampling time of the weighing data; This indicates the position of the disk corner at the current visual data sampling moment; This represents the fixed mechanical installation angle difference, used to characterize the theoretical angular displacement relationship; It represents the angular displacement fluctuation value, which is used to measure the normal scale of spatial deviation and to normalize the angle term; This indicates a preset positive number used to avoid zero denominators and ensure calculation stability; the preferred value is [value missing]. arrive .

[0034] In this implementation scheme, by uniformly normalizing the time deviation and angular displacement deviation and constructing a correlation residual criterion that integrates the two, the consistency assessment of visual data and weighing data in the spatiotemporal dimension is achieved. The original scattered matching process is transformed into a single quantitative indicator. Thus, even in the presence of cycle fluctuations, mechanical errors and noise interference, the weighing data that best matches the current visual data can still be accurately selected, effectively reducing the probability of mismatch across material boxes and improving the data binding accuracy and operational stability.

[0035] Specifically, the process of binding multi-source data based on the association matching results to obtain fused data is as follows: In the candidate weighing data set corresponding to the current visual data, the corresponding association residual value is calculated for each weighing data, and the weighing data with the smallest association residual value is selected as the optimal matching result. The minimum association residual principle is used to achieve optimal consistency screening of candidate weighing data in the time and space dimensions, thereby ensuring that the selected data has the highest degree of matching with the current visual data under multiple feature constraints. The stable weight value, weighing acquisition timestamp, and disk corner position corresponding to the optimally matched weighing data are bound to the current visual data to generate a bin-level visual weight alignment data record, thus obtaining the bin's fused data. The minimum association residual value is also recorded as the matching confidence index of the fused data, which is used to characterize the reliability of the current binding result. Preferably, after selecting the optimal matching result, a threshold check is performed on the minimum correlation residual value. When the minimum correlation residual value is greater than the residual threshold, it is determined that the current matching result has an unreliable risk, and a secondary check mechanism is triggered. The secondary check mechanism includes re-acquiring the weighing data in the subsequent time window as an expanded candidate set, thereby avoiding mismatches caused by cycle fluctuations and abnormal disturbances. When the minimum correlation residual value is less than or equal to the residual threshold, the current matching result is maintained as a valid binding result.

[0036] In this implementation scheme, the optimal matching of visual data and weighing data is achieved based on the principle of minimum correlation residual. On this basis, a residual threshold verification and secondary verification mechanism are introduced to determine the reliability of the matching results and correct anomalies. This effectively avoids misbinding in the presence of beat fluctuations and noise interference, forming a closed-loop control for data matching and improving the accuracy of multi-source data fusion and the stability of system operation.

[0037] Specifically, based on fused data, the process of analyzing appearance integrity using the effective target area area and defect area area, analyzing weight consistency using stable weight values, and combining appearance integrity and weight consistency to determine the overall state of the material box for anomaly judgment is as follows: For each material box, the corresponding fused data is read, and according to the product category to which the material box belongs, the reference effective area area and reference weight value of the corresponding category are read; wherein the reference effective area area and reference weight value are preferably obtained by selecting the corresponding category of material boxes and filtering the material box data judged as normal in the historical sorting results, and calculating the statistical median value of the effective target area area and stable weight value as reference values ​​respectively; in the initial operation stage, the reference effective area area and reference weight value are preferably obtained through preset calibration samples, and dynamically updated in subsequent operation in combination with sorting results to ensure that the reference values ​​can reflect the true distribution characteristics of the current production state, thereby maintaining consistency with the feedback update mechanism in the sorting decision module. Divide the current stable weight value of the hopper by the reference weight value of the corresponding category to obtain the weight ratio. The weight ratio is used to characterize the degree of deviation of the current hopper's actual weight from the standard weight, thereby reflecting whether there is a shortage, over-filling, or mixed material situation in the hopper's loading status. Add a preset positive number to the weight ratio to obtain the weight item. By introducing a preset positive number, the denominator is avoided from being zero and the logarithm is undefined in subsequent calculations, thus ensuring the stability of the calculation process. Divide the effective target area area of ​​the current material box by the reference effective area area of ​​the corresponding category to obtain the area ratio. The area ratio characterizes how close the visual occupancy of the current material box is to the standard state. Divide the defect area area of ​​the current material box by the sum of the effective target area area and a preset positive number to obtain the defect percentage value. The defect percentage value quantifies the proportion of abnormal parts in the target area of ​​the material box, thus reflecting the severity of appearance defects. Subtract the defect percentage value from one to obtain the non-defect percentage value. Multiply the area ratio by the non-defect percentage value to obtain the visual comprehensive percentage value. The non-defect percentage value characterizes the integrity of the normal area in the effective area of ​​the material box. Take the square root of the visual comprehensive percentage value and add the preset positive number to obtain the visual item. The preset positive number is used to avoid the denominator and logarithm input being zero, ensuring calculation stability. The preferred value is [value missing]. arrive The visual features are scaled using square root operations to reduce the excessive influence of local anomalies on the overall evaluation and improve the smoothness of the visual evaluation. The visual weight consistency deviation value of the material box is obtained by dividing the weight term by the visual term, taking the natural logarithm, and then taking the absolute value of the natural logarithm result. The ratio of the weight term to the visual term characterizes the degree of consistency between the weight and visual states; the natural logarithm operation compresses the range of the ratio, making different degrees of deviation comparable on a uniform scale; the absolute value operation eliminates the influence of the deviation direction, making weight overestimation and underestimation equivalent in the evaluation, thus constructing a symmetrical consistency evaluation index. The final visual weight consistency deviation value comprehensively reflects the matching degree of the material box in both visual and weight dimensions. A larger value indicates increased inconsistency between the two, which can be used to identify latent or compound anomalies.

[0038] The specific formula for the visual weight consistency deviation value is as follows: ; In the formula, The visual weight consistency deviation value is used to comprehensively evaluate the consistency between the visual state and the weight state of the material box. It is obtained by comparing the degree of deviation of the weight from the standard value with the degree of visual integrity. The larger the deviation value, the more inconsistent the visual and weight states are, and it is used to identify hidden abnormalities or abnormal materials. It represents a stable weight value, reflecting the actual loaded mass of the material box; This indicates a reference weight value, which serves as a standard weight benchmark. This indicates the effective target area area, reflecting the extent to which the material box effectively occupies visual space; This indicates the area of ​​the effective reference region, which serves as the standard visual area benchmark. This indicates the area of ​​the defective region, used to characterize the degree of appearance abnormality; This indicates a preset positive number, used to avoid inputting zero in the denominator or logarithm, ensuring calculation stability. The preferred value is [value missing]. arrive .

[0039] In this embodiment, Table 1 is a data table of visual weight consistency deviation values. The reference weight value is 50, the effective reference area is 12000, and the preset positive number is... The table details the stable weight, effective target area, defect area, and visual weight consistency deviation for different types of material boxes of the same type. Specifically, material box 1 has a stable weight of 49.8, an effective target area of ​​11850, a defect area of ​​0, and a visual weight consistency deviation of 0.0023; material box 2 has a stable weight of 45.2, an effective target area of ​​11780, a defect area of ​​0, and a visual weight consistency deviation of 0.0917; material box 3 has a stable weight of 54.6, an effective target area of ​​11998, a defect area of ​​0, and a visual weight consistency deviation of 0.0881; material box 4 has a stable weight of 49.3, an effective target area of ​​11050, a defect area of ​​1200, and a visual weight consistency deviation of 0.0846; and material box 5 has a stable weight of 47.5, an effective target area of ​​11200, a defect area of ​​600, and a visual weight consistency deviation of 0.0107.

[0040] Table 1. Data on Visual Weight Consistency Deviation

[0041] like Figure 3 The diagram shows the distribution of visual weight consistency deviation values ​​for different material boxes. It illustrates the visual weight consistency deviation values ​​for different material boxes, with dashed lines indicating deviation thresholds to distinguish between normal and abnormal boxes. The horizontal axis represents the box number, and the vertical axis represents the visual weight consistency deviation value. A larger bar indicates a greater deviation between visual and weight measurements; when the deviation threshold is exceeded, the material box is considered abnormal. (Refer to Table 1 and...) Figure 3 It can be seen that the visual weight consistency deviation values ​​of different boxes show significant differences. Boxes 1 and 5 have deviation values ​​significantly lower than the deviation threshold, indicating good matching between visual and weight characteristics and high consistency, thus they can be classified as normal boxes. Although box 5 has a defective area, it does not exceed the defect area threshold, and even minor flaws can be classified as normal boxes. Boxes 2 and 3 have deviation values ​​significantly higher than the threshold, with effective target area areas close to the reference value and corresponding defect area areas of zero, indicating that their appearance is normal but their weight is abnormal, belonging to hidden abnormal boxes that are difficult to identify visually. Box 4 also has a deviation value higher than the threshold, and the defect area is large, indicating the simultaneous presence of visual defects and consistency deviation issues. Further comparison shows that when the visual change of a box is in the same direction as the weight change, its deviation value is small, while when the visual state and weight state do not match, its deviation value increases significantly. This verifies that the method can effectively identify abnormal situations where visual and weight are inconsistent, especially suitable for detecting boxes with normal appearance but internal abnormalities, improving the accuracy and reliability of sorting judgment.

[0042] In this implementation plan, a reference effective area and reference weight value are constructed based on historical normal samples to achieve standardized comparison of visual and weight characteristics of the material box. A comprehensive evaluation index for appearance integrity and loading status is constructed by using area ratio, defect ratio, and weight ratio. Furthermore, by performing scale compression processing on visual features and introducing logarithmic compression and symmetry processing mechanisms, the consistent relationship between visual status and weight status is transformed into a single quantitative index. This enables stable and comparable consistency evaluation under different working conditions, effectively identifying hidden problems and compound anomalies where the appearance is normal but the weight is abnormal, reducing the misjudgment rate and improving the accuracy and reliability of sorting judgment.

[0043] Specifically, the process of judging the quality of the material box based on the anomaly judgment result is as follows: The visual weight consistency deviation value of the material box is compared with the deviation threshold to judge the quality. The deviation threshold is preferably obtained based on the statistical analysis of historical normal material box samples of the corresponding product category. It is obtained by performing distribution analysis on the visual weight consistency deviation value of historical normal samples, so that the threshold can reflect the normal fluctuation range and adapt to the characteristic differences of different categories of material boxes. In the initial stage, the deviation threshold can be set by preset calibration samples. When the visual weight consistency deviation value is greater than the deviation threshold, the material box is judged as an abnormal material box. When the visual weight consistency deviation value is less than or equal to the deviation threshold, the material box is judged as a normal material box. At the same time, the defect area area of ​​the corresponding material box is read. If the defect area area is greater than the defect area threshold, the material box is judged to have an appearance abnormality and the material box is directly marked as an abnormal material box. The defect area threshold is preferably determined based on the range of minor defects allowed in a normal material box. This is used to distinguish between acceptable slight apparent fluctuations and actual abnormal defects, thereby avoiding misjudgments caused by noise and minor disturbances. By combining the judgment of visual weight consistency deviation with the judgment of defect area, the state of the material box can be constrained from two dimensions: overall consistency and local defects. This enables the collaborative identification of latent and explicit abnormalities, improving the comprehensiveness and accuracy of abnormality judgment.

[0044] In this implementation scheme, by jointly judging the deviation value of visual weight consistency and the area of ​​defective region, the overall consistency abnormality and local appearance abnormality of the material box can be collaboratively identified. The judgment threshold obtained based on the statistics of historical normal samples makes the judgment standard adaptive and stable, thereby effectively distinguishing normal fluctuations from actual abnormalities under different working conditions, reducing the false judgment rate and improving the accuracy of material box quality judgment and the reliability of system operation.

[0045] Specifically, the process of generating sorting decisions based on the quality judgment results of the material boxes, controlling the material boxes to perform classification and sorting at downstream workstations, and providing feedback on the sorting results is as follows: After the material boxes complete the quality judgment, they are transferred from the disc-carrying workstation to the downstream conveying path. The transfer process is realized through a synchronous conveying mechanism, ensuring that the material boxes smoothly transition from the disc-carrying workstation to the downstream conveying path according to a predetermined rhythm, thereby ensuring the stability of the material boxes' posture and controllable position during the transfer process. When the material boxes reach the sorting workstation, sorting control commands are generated. The sorting control commands are generated in real time by the PLC based on the quality judgment results of the material boxes and the current conveying rhythm, and sent to the sorting execution mechanism through the industrial communication interface: When a material box is judged to be a normal material box... During sorting, the control sorting mechanism transports the boxes to the target grid in the preset multi-station sorting grid. The target grid is configured according to production needs and batch rules to achieve classified storage of boxes of different categories and batches. When a box is determined to be an abnormal box, the control sorting mechanism transports the box to the waste hopper to automatically remove the abnormal box and prevent it from entering subsequent processes. After sorting, the sorting execution time, sorting execution status, and sorting destination of the current box are recorded and associated with the fused data of the corresponding box to form a box sorting result record. The sorting execution status includes sorting success, execution failure, and abnormal interruption information for subsequent system operation monitoring and fault tracing. Based on the fused data of the sorted boxes, the reference weight value, reference effective area area, typical migration time, time fluctuation value, and angular displacement fluctuation value are updated through statistical update. The reference weight value and reference effective area area are only selected from boxes with normal sorting results for updating. By dynamically correcting the above parameters, we can continuously adapt to changes in the production environment, improve the accuracy and stability of parameter estimation, and avoid abnormal samples from interfering with the statistical results.

[0046] In this implementation plan, by linking the quality judgment results with the sorting execution, the automatic classification and rejection of abnormalities in the material boxes are realized. The key reference parameters are dynamically updated in combination with the sorting results, which can continuously adapt to changes in production cycle and equipment operating status. This ensures sorting accuracy while improving the stability of parameter estimation and the reliability of long-term operation.

[0047] Reference Figure 2As shown, the second aspect of the present invention provides a box sorting data processing system based on visual weight fusion, applied to the aforementioned box sorting data processing method based on visual weight fusion, comprising: a sorting data acquisition and processing module, used to acquire visual data, weighing data, and corresponding position and time information of boxes in the disc line in real time, perform unified time base alignment and preprocessing on multi-source data, extract the appearance area features and stable weight values ​​of the boxes, and form a standardized box feature set, wherein the stable weight value is used to reflect the actual loading mass of the boxes; and a box trajectory alignment module, used to analyze the motion trajectory relationship of the boxes on the disc using the standardized box feature set, obtain time features and spatial position features, and combine the time features and spatial position features. The system uses a feature-based correlation matching method to associate visual data and weighing data, and binds multi-source data based on the correlation matching results to obtain fused data. The visual weight fusion evaluation module analyzes appearance integrity using the effective target area area and defect area area based on the fused data, analyzes weight consistency using stable weight values, and combines appearance integrity and weight consistency to determine anomalies in the overall state of the box. Based on the anomaly determination results, the quality of the box is judged. The effective target area area is used to reflect the effective occupancy of the box visually. The sorting decision feedback module generates sorting decisions based on the box quality judgment results, controls the boxes to perform classification and sorting at downstream workstations, and provides feedback on the sorting results.

[0048] like Figure 4 The diagram shows the actual operation of a box sorting system based on visual-weight fusion. Centered on a disc-mounted workstation, multiple workstations (L1-L8) are set up on the disc to carry boxes, which then rotate through each functional workstation sequentially. Boxes are input from the loading station and placed into the disc-mounted workstation. During the disc's rotation, the boxes first pass through a visual inspection station to acquire image information about their appearance. They then continue rotating to the weighing station, where they are continuously weighed as they pass through the weighing area, obtaining the corresponding weight data. During this process, the encoder synchronously outputs the disc's corner position, achieving continuous spatial calibration of the boxes and establishing a temporal and spatial correspondence between visual and weighing data. After inspection, the boxes are output to the downstream sorting station via a transfer station. Based on the aforementioned visual-weight fusion evaluation results, they are classified and sorted. Boxes deemed normal are transported to their target compartments, while those deemed abnormal are transported to the waste bin. Through the above structural layout and operation process, continuous detection of the material box during the disc movement process, multi-source data acquisition, and spatiotemporal correlation-based fusion processing are realized, providing reliable data support for subsequent quality judgment and sorting decisions.

[0049] In this implementation plan, a complete closed-loop processing flow is constructed, from multi-source data acquisition, spatiotemporal alignment, fusion evaluation to sorting decision and feedback update. This enables collaborative analysis and consistency judgment of visual and weight information, and can accurately identify hidden anomalies and explicit defects in the material box under complex working conditions. At the same time, the dynamic parameter update mechanism enables the system to continuously adapt to production changes, thereby improving the accuracy, stability and overall reliability of sorting judgment.

[0050] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0051] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. As those skilled in the art will understand, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A data processing method for box sorting based on visual weight fusion, characterized in that, Includes the following steps: S1, real-time acquisition of visual data, weighing data and corresponding position and time information of the material box in the disc line, unified time base alignment and preprocessing of multi-source data, extraction of material box appearance area features and stable weight value, forming a standardized material box feature set, the stable weight value is used to reflect the actual loading mass of the material box; S2, using the standardized material box feature set to analyze the motion trajectory relationship of the material box on the disc, obtain the temporal and spatial features, combine the temporal and spatial features to perform correlation matching on visual data and weighing data, and bind multi-source data according to the correlation matching results to obtain fused data; The specific process of combining temporal and spatial features to correlate and match visual data and weighing data is as follows: For each weighing data in the candidate weighing data set, the actual time difference is obtained by subtracting the current visual acquisition timestamp from the corresponding weighing acquisition timestamp. The time deviation value is obtained by subtracting the typical migration time from the actual time difference. The normalized time deviation is obtained by dividing the time deviation value by the sum of the time fluctuation value and the preset positive number. Calculate the difference between the disk angle position corresponding to the sampling time of the weighing data and the disk angle position corresponding to the sampling time of the current visual data to obtain the actual angular displacement difference. Subtract the fixed mechanical installation angle difference from the actual angular displacement difference to obtain the angular displacement deviation value. Divide the angular displacement deviation value by the sum of the angular displacement fluctuation value and the preset positive number to obtain the normalized angular displacement deviation amount. The square root of the sum of the squares of the normalized time deviation and the normalized angular displacement deviation is taken to obtain the correlation residual between the visual data and the candidate weighing data. The specific process of binding multi-source data based on the association matching results to obtain fused data is as follows: In the candidate weighing data set corresponding to the current visual data, the corresponding correlation residual value is calculated for each weighing data, and the weighing data with the smallest correlation residual value is selected as the optimal matching result; the stable weight value, weighing acquisition timestamp and disk corner position corresponding to the optimal matching weighing data are bound to the current visual data to generate a hopper-level visual weight alignment data record, and the fusion data of the hopper is obtained. S3. Based on the fused data, analyze the appearance integrity using the effective target area area and the defect area area, analyze the weight consistency using the stable weight value, combine the appearance integrity and weight consistency to make anomaly judgment on the overall state of the material box, and make a quality judgment on the material box based on the anomaly judgment result. The effective target area area is used to reflect the effective occupancy of the material box in vision. S4 generates sorting decisions based on the quality judgment results of the material boxes, controls the material boxes to perform classification and sorting at downstream workstations, and provides feedback on the sorting results.

2. The data processing method for box sorting based on visual weight fusion according to claim 1, characterized in that, The specific process of real-time acquisition of visual data, weighing data, and corresponding position and time information of the material box in the disc production line is as follows: As the material box rotates with the disc and passes through each inspection station in sequence, the sorting data of the material box is collected in real time, including: using an industrial camera at the vision inspection station to collect images of the passing material boxes to obtain visual data, including the image frame of the corresponding material box and the vision acquisition timestamp; using a weighing sensor at the weighing station to continuously weigh the passing material boxes to obtain weighing data, including the weighing value and the weighing acquisition timestamp; and at the same time, the encoder continuously samples during the disc rotation process to obtain the disc corner position corresponding to each sampling moment.

3. The data processing method for box sorting based on visual weight fusion according to claim 1, characterized in that, The specific process of performing unified time-base alignment and preprocessing on multi-source data, extracting the appearance area features and stable weight values ​​of the utensil box, and forming a standardized utensil box feature set is as follows: The visual acquisition timestamp and the weighing acquisition timestamp are uniformly mapped to the standard clock reference of the production line for time alignment. Gaussian filtering is used to denoise the image frame, and histogram equalization is used to perform gray-level equalization on the denoised image frame. Based on threshold segmentation and connected component analysis, the region where the utensil is located in the image frame is located to obtain the pixel set of the target region of the utensil. The effective target region area of ​​the utensil is calculated by counting the number of pixels in the set. Within the target region of the utensil, abnormal regions are identified by segmentation methods based on gray-level differences and edge features. The pixel set of the abnormal regions is determined as the defect region pixel set, and the defect region area is calculated by counting the number of pixels in the set. The difference between the weighing values ​​of adjacent sampling points is calculated to obtain the weighing change. When the weighing change of two consecutive sampling points is greater than the judgment threshold, the corresponding sampling point is marked as a candidate position for the weighing start point. When the weighing change of two consecutive sampling points is less than the negative of the judgment threshold, the corresponding sampling point is marked as a candidate position for the weighing end point. Between the candidate positions for the weighing start point and the candidate positions for the weighing end point, a sampling segment is extracted where the weighing value is continuously higher than the empty load threshold and the number of consecutive sampling points meets the minimum sampling point threshold. This segment is used as the weighing data segment corresponding to the material box. Within the weighing data segment corresponding to each material box, median filtering is performed on the weighing value sequence. Based on the robust anomaly detection method of median and median absolute deviation, abnormal weighing values ​​are identified and removed. The median of the remaining weighing values ​​is taken as the stable weight value of the material box. Normalization processing is performed on the material box sorting data, and a material box sorting database is established to store the original and preprocessed material box sorting data.

4. The method for processing box sorting data based on visual weight fusion according to claim 1, characterized in that, The specific process of analyzing the motion trajectory relationship of the material box on the disk using a standardized material box feature set to obtain the temporal and spatial position features is as follows: For each visual data point, read the visual acquisition timestamp and the corresponding disk corner position at the sampling time, and calculate the difference between the current visual acquisition timestamp and each weighing acquisition timestamp. Select weighing data points with differences less than the time difference threshold as candidate weighing data sets. Read the weighing acquisition timestamp and the corresponding disk corner position at the sampling time for each weighing data point in the candidate weighing data set. Based on the known structural conditions of the disk equipment, obtain the fixed mechanical installation angle difference between the visual inspection station and the weighing station. Based on the historically bound and sorted boxes, the time difference sequence between the weighing sampling timestamp and the visual sampling timestamp is statistically analyzed. The median value of the time difference sequence is taken as the typical migration time, and the median absolute deviation of the time difference sequence is calculated as the time fluctuation value. At the same time, the difference sequence of the disk corner position from the historical visual inspection station to the weighing station is statistically analyzed, and the median absolute deviation is calculated to obtain the angular displacement fluctuation value.

5. The method for processing box sorting data based on visual weight fusion according to claim 1, characterized in that, The specific process of analyzing appearance integrity using effective target area and defect area based on fused data, analyzing weight consistency using stable weight values, and combining appearance integrity and weight consistency to determine anomalies in the overall state of the material box is as follows: For each material box, read the corresponding fusion data, and according to the product category to which the material box belongs, read the reference effective area area and reference weight value of the corresponding category; Divide the current stable weight value of the material box by the reference weight value of the corresponding category to obtain the weight ratio. Add a preset positive number to the weight ratio to obtain the weight item. Divide the effective target area of ​​the current material box by the reference effective area of ​​the corresponding category to obtain the area ratio; divide the defect area of ​​the current material box by the sum of the effective target area and the preset positive number to obtain the defect percentage; subtract the defect percentage from one to obtain the non-defect percentage; multiply the area ratio by the non-defect percentage to obtain the visual comprehensive percentage; take the square root of the visual comprehensive percentage and add the preset positive number to obtain the visual item. The visual weight consistency deviation of the box is obtained by dividing the weight item by the visual item and taking the natural logarithm, and then taking the absolute value of the natural logarithm result.

6. The method for processing box sorting data based on visual weight fusion according to claim 1, characterized in that, The specific process for determining the quality of the material box based on the anomaly detection result is as follows: The visual weight consistency deviation of the material box is compared with the deviation threshold to determine the quality: when the visual weight consistency deviation is greater than the deviation threshold, the material box is determined to be an abnormal material box; when the visual weight consistency deviation is less than or equal to the deviation threshold, the material box is determined to be a normal material box; at the same time, the defect area of ​​the corresponding material box is read. If the defect area is greater than the defect area threshold, the material box is determined to have an appearance abnormality and the material box is directly marked as an abnormal material box.

7. The method for processing box sorting data based on visual weight fusion according to claim 1, characterized in that, The specific process of generating sorting decisions based on the quality judgment results of the material boxes, controlling the material boxes to be classified and sorted at downstream workstations, and providing feedback on the sorting results is as follows: After the material box completes the quality judgment, the material box is transferred from the disc carrier station to the downstream conveying path. When the material box runs to the sorting station, a sorting control command is generated: when the material box is judged to be a normal material box, the sorting execution mechanism is controlled to transport the material box to the target grid in the preset multi-station sorting grid; when the material box is judged to be an abnormal material box, the sorting execution mechanism is controlled to transport the material box to the waste hopper. After sorting is completed, the sorting execution time, sorting execution status and sorting destination of the current box are recorded and associated with the fusion data of the corresponding box to form a box sorting result record. Based on the fusion data of the sorted boxes, the reference weight value, reference effective area, typical migration time, time fluctuation value and angular displacement fluctuation value are updated.

8. A box sorting data processing system based on visual weight fusion, employing the box sorting data processing method based on visual weight fusion as described in any one of claims 1-7, characterized in that, include: The sorting data acquisition and processing module is used to collect visual data, weighing data and corresponding position and time information of the boxes in the disc line in real time. It performs unified time base alignment and preprocessing on multi-source data, extracts the appearance area features and stable weight values ​​of the boxes, and forms a standardized box feature set. The stable weight value is used to reflect the actual loading mass of the boxes. The bin trajectory alignment module is used to analyze the motion trajectory relationship of the bin on the disc using a standardized bin feature set, obtain temporal and spatial features, combine the temporal and spatial features to perform correlation matching between visual data and weighing data, and bind multi-source data according to the correlation matching results to obtain fused data. The visual weight fusion assessment module is used to analyze appearance integrity based on fused data, using the area of ​​effective target area and the area of ​​defective area, analyze weight consistency using stable weight value, combine appearance integrity and weight consistency to make anomaly judgment on the overall state of the box, and make box quality judgment based on the anomaly judgment result. The area of ​​effective target area is used to reflect the effective occupancy of the box in vision. The sorting decision feedback module is used to generate sorting decisions based on the quality judgment results of the material boxes, control the material boxes to perform classification and sorting at downstream workstations, and provide feedback on the sorting results.

Images