Plastic tarpaulin quality comprehensive evaluation method based on deep learning

Abstract

The present application relates to the technical field of plastic tarpaulin quality evaluation, in particular to a plastic tarpaulin quality comprehensive evaluation method based on deep learning, which comprises: extracting defect features from production line image blocks and constructing a distribution atlas set; comparing feature response differences and generating key quality monitoring area bands according to texture connection relationships; extracting process parameter fluctuation trends and combining process identifiers to generate production quality correlation interval sequences; extracting position-coincident process segments and combining visual and physical features to generate a quality evaluation candidate feature set; and superimposing multi-dimensional features to compress quality change lines and fit measured indicators to generate a comprehensive evaluation report. The present application can establish a stable association between production trajectories and appearance defects, effectively enhance the matching completeness and evaluation objectivity of visual information and measured data in quality determination through a unified evaluation structure of multi-source monitoring data.

Description

Technical Field

[0001] This invention relates to the field of plastic tarpaulin quality assessment technology, and in particular to a comprehensive quality assessment method for plastic tarpaulins based on deep learning. Background Technology

[0002] The field of plastic tarpaulin quality assessment technology encompasses functions such as material performance testing, appearance defect identification, structural stability analysis, processing consistency assessment, and finished product quality grading. Its core content involves centrally describing the quality status formed during stages such as raw material proportioning, extrusion film formation, weaving and lamination, coating and bonding, thermal bonding, and finished product inspection within a unified quality evaluation system through indicator collection, comparison, and judgment. This generates quality analysis reports on thickness uniformity, surface smoothness, tensile properties, tear resistance, weather resistance, and waterproof performance. This field systematically includes data acquisition (such as surface image acquisition, production process recording, and test data entry), data preprocessing (such as image annotation, indicator quantification, and noise reduction), data feature extraction (such as convolution feature extraction and defect area identification), data analysis (such as multi-indicator correlation analysis and weighted comparison of single results), and output of finished product grade determination and assessment services.

[0003] Among them, the deep learning-based comprehensive quality assessment method for plastic tarpaulins addresses technical issues in the plastic tarpaulin quality assessment process, including the identification of appearance defects, integration of physical performance indicators, unified judgment of test results, and comprehensive analysis of multiple quality information. Specifically, it involves preprocessing data from various sources (e.g., labeling defect samples on tarpaulin surface images, normalizing thickness and weight data, and pre-classifying quality categories), then identifying and associating these data (e.g., identifying defect areas such as scratches, holes, and wrinkles, associating tensile strength, tear strength, and peel strength, and statistically analyzing waterproof value and folding resistance), and finally using a deep learning mechanism to adjust the weights or assessment strategies of each test item, enabling the assessment model to adaptively select the quality criteria and assessment logic for judgment.

[0004] Existing technologies rely on manual sampling or single physical index measurement. Under conditions of multi-process composite or complex appearance and texture, it is difficult to effectively express the coupling law between surface defect distribution and internal performance degradation. There is a lack of anomaly identification mechanism based on deep feature aggregation, making it difficult to construct continuous quality spatial correlation. Inspection records are only used for static index alignment and do not combine the fluctuation trend of production process to establish quality stability judgment. This results in a lack of dynamic correspondence between process monitoring records and finished product performance data. The evaluation content is mostly performed on single-item judgment and lacks a fine fusion method for multi-source inspection indicators. There is a logical deviation between visual inspection information and physical measurement data in detail judgment, which affects the classification accuracy and objective integrity of the final evaluation results. Summary of the Invention

[0005] To address the technical problems existing in the prior art, embodiments of the present invention provide a comprehensive quality evaluation method for plastic tarpaulins based on deep learning. The technical solution is as follows: A deep learning-based comprehensive quality assessment method for plastic tarpaulins includes the following steps: S1: Obtain image records of appearance defects formed on the plastic tarpaulin production line, divide them into continuous image blocks according to a fixed sampling window, extract the defect feature positions in each image block and rearrange them according to row and column coordinates to form block-level quality description results, and connect all block-level quality description results in the production material flow sequence to generate a defect distribution feature map set. S2: Based on the defect distribution feature map set, compare the feature response differences of adjacent image blocks, filter out image groups whose response values ​​exceed the quality discrimination threshold, organize them into continuous regions according to the physical connection relationship of the tarpaulin surface texture, mark the continuous regions with abnormal associations, and generate key quality monitoring area zones. S3: Acquire the physical performance monitoring records collected by sensors during the production process and connect the process trajectory in series according to the production batch time. Extract the production segments with consistent fluctuation trends of process parameters and overlay the corresponding production process identifiers. Integrate the process position of the production segment with the spatial position of the key quality monitoring area to generate a production quality correlation interval sequence. S4: Based on the production quality correlation interval sequence and key quality monitoring area band, extract the process monitoring segments with overlapping positions, summarize the corresponding surface image features and physical performance measurement content one by one, combine the monitoring content at the same position into evaluation basic items, and generate a quality assessment candidate feature set.

[0006] As a further aspect of the present invention, the defect distribution feature map set includes defect pixel coordinate distribution, image block sequence, and defect type index; the key quality monitoring area band includes continuous abnormal area identifiers, evaluation priority markers, and process matching sequences; the production quality correlation interval sequence includes process parameter fluctuation information, continuous batch numbers, and physical index correspondence; and the quality assessment candidate feature set includes an evaluation item index, visual and physical data pairing results, and a list of overlapping process segments.

[0007] As a further aspect of the present invention, the step of obtaining S1 is as follows: S101: Acquire surface image records of plastic tarpaulin formed during the operation of the collection equipment on the production line, and locate the pixel grayscale data contained in each frame of the image in sequence according to the camera trigger scanning order. According to the preset detection area mask configuration parameter set, the image frame is cropped within the effective detection range, and split into multiple image block segments in the continuous acquisition order. The surface position index data of each segment is numbered to obtain the image position cropping sequence. S102: Based on the surface index range of each image block segment in the image position cropping sequence, extract the feature response value of each pixel in the segment, identify the feature dense point group located in the local area, perform response sequence update operation on each group of pixel coordinates by comparing the horizontal and vertical offset relationship, and output the feature point position structure. S103: The coordinate sequence of each image block segment in the feature point positional structure is called to perform a time-series connection operation according to the acquisition timestamp order, and the feature sequences corresponding to multiple segments are connected to the continuous detection channel in chronological order to obtain a defect distribution feature map set.

[0008] As a further aspect of the present invention, the step of obtaining S2 is as follows: S201: Based on the adjacent segment data in the defect distribution feature map set, the corresponding feature point information in each segment is compared in sequence, the difference in feature response values ​​in the horizontal and vertical directions is retrieved, and the difference is confirmed item by item with the preset quality offset comparison range. All segments with differences exceeding the quality discrimination threshold are screened out to obtain the abnormal feature index column. S202: Call the fragment number in the abnormal feature index column, extract the data frame of adjacent number according to the production direction, identify all fragment combinations with consecutive numbers and located in the same detection area, determine whether the number of consecutive occurrences in the sequence exceeds the regional connectivity benchmark value, output the number set for fragment combinations that meet the sequence continuity rule, and obtain the abnormal fragment chain set list. S203: Based on the numbered data in the abnormal fragment chain set list, perform a physical position matching operation on the area corresponding to each chain in the tarpaulin surface texture data, and group the continuously connected fragment chains according to the surface warp and weft structure connection relationship, and assign key attribute identification information to obtain the key quality monitoring area zone.

[0009] As a further aspect of the present invention, the step of obtaining S3 is as follows: S301: Acquire the physical performance monitoring records generated during the operation of the plastic tarpaulin production equipment, and arrange all records in chronological order into a process data frame sequence. Extract parameter fluctuation information for each record in the sequence and perform trend judgment operation on the direction change of the fluctuation vector. By comparing the stability of the slope sign between adjacent fluctuation vectors, confirm the process segments with consistent trends, and add corresponding process position identification codes to these segments to obtain the trend identification process segment column. S302: Call the process position identifier code in the trend identifier process segment column, sequentially retrieve the code data of each process segment, extract the continuously numbered process segments according to the arrangement order of the process segments in the production line, confirm the set of process segments whose number change range is within the process continuity reference value range through the sequence continuity judgment operation, and output the process position sequence of these process segments to obtain the continuous production process space column. S303: Based on the process position sequence in the continuous production process space column, and the process position data of each segment in the critical quality monitoring area, perform a one-to-one comparison operation on the coordinates of the two under the same production line index system, extract the position segment of the continuous process space sequence in the critical quality monitoring area from the correspondence, and arrange the output sequence according to the production order to obtain the production quality related interval sequence.

[0010] As a further aspect of the present invention, the step of obtaining S4 is as follows: S401: Based on the production quality correlation interval sequence and the key quality monitoring area, call the process position index information of each segment in the production quality correlation interval, compare its position number with the position number in the key quality monitoring area under the production line index system, perform a screening operation on the data frames with the same coordinate number, extract the segment content with overlapping process positions, and obtain a list of overlapping process segments. S402: Based on the set of numbers in the list of overlapping process segments, extract the visual description content associated with the corresponding segment in the surface image feature set and the measured index content corresponding to the physical index dataset in sequence. Aggregate the two types of monitoring content with the same process index in numerical order and compress them into an evaluation item structure to obtain the visual dual-mode evaluation item set. S403: Call the process number information in the visual dual-mode evaluation item set, arrange all evaluation item structures according to the production flow index order, organize the item sequence according to the number change trend, output the evaluation item dataset with continuous process number characteristics, and obtain the quality evaluation candidate feature set.

[0011] As a further aspect of the present invention, the method further includes: S5: Call the visual description content and measured index content in the quality assessment candidate feature set, superimpose and compress multiple visual feature descriptions into a single quality change line at the same process position, attach the measured index content to the quality change line at the corresponding position, rearrange all quality segments according to the production sequence, and generate a comprehensive tarpaulin quality assessment report. The comprehensive quality assessment report for the tarpaulin includes the compression feature line structure, the bonding performance index layer, and the unified process assessment segment.

[0012] As a further aspect of the present invention, the step of obtaining S5 is as follows: S501: Call the visual description content in the quality assessment candidate feature set, extract multidimensional response values ​​for the feature line data of all items under the same process number in numerical order, and sequentially stack and process the quality evolution paths of all co-position line segments along the production direction, compress to form an assessment unit structure, and obtain a compressed quality evolution line set. S502: Based on the path location index in the compressed quality evolution line set, call the measured index content under the corresponding number, locate the index change limit value on both sides of each path, and perform close-fitting processing on each index boundary and quality evolution path line, output the combined item structure composed of index boundary and quality line, and obtain the performance fit evaluation item list. S503: Call the number sequence information in the performance fit evaluation item column, extract all combined items according to the continuous process direction, construct a continuous quality evaluation structure according to the arrangement order of the numbers in the production axis, integrate all combined items to form a continuous sequence, and obtain a comprehensive quality evaluation report of tarpaulin.

[0013] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this invention, by dividing tarpaulin image data into continuous image blocks according to the acquisition sequence and constructing a distribution map, a set of defect information with sequential characteristics can be formed in the surface structure. By comparing response differences and combining abnormal fragment chains, a quality fluctuation sequence with production logic is constructed. By corresponding process parameter trends with process positions, a stable correlation between production trajectory and appearance defect bands is established. By combining visual features and physical indicators of the evaluation position, evaluation items with synchronous process benchmarks are generated. By compressing the quality evolution line and attaching physical indicators, multi-source monitoring data are linked to form a unified evaluation structure, which enhances the matching completeness and evaluation objectivity of visual information and physical measured data in the quality judgment process. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart illustrating the acquisition process of S1 in this invention; Figure 3 This is a flowchart illustrating the acquisition process of S2 in this invention; Figure 4 This is a flowchart illustrating the acquisition process of S3 in this invention; Figure 5 This is a flowchart illustrating the acquisition process of S4 in this invention; Figure 6 This is a flowchart of the acquisition process for S5 of the present invention. Detailed Implementation

[0015] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0016] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0017] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0018] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0019] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0020] Please see Figure 1 This invention provides a technical solution: a comprehensive quality evaluation method for plastic tarpaulins based on deep learning, comprising the following steps: S1: Obtain image records of appearance defects formed on the plastic tarpaulin production line, divide them into continuous image blocks according to a fixed sampling window, extract the defect feature positions in each image block and rearrange them according to row and column coordinates to form block-level quality description results, and connect all block-level quality description results in the production material flow sequence to generate a defect distribution feature map set. S2: Based on the difference in feature response of adjacent image blocks compared with the defect distribution feature map set, image groups whose response values ​​exceed the quality discrimination threshold are selected, and the continuous regions are organized according to the physical connection relationship of the tarpaulin surface texture. The continuous regions are marked with abnormal association markers to generate key quality monitoring area zones. S3: Acquire the physical performance monitoring records collected by sensors during the production process and connect the process trajectory in series according to the production batch time. Extract the production segments with consistent fluctuation trends of process parameters and overlay the corresponding production process identifiers. Integrate the process position of the production segment with the spatial position of the key quality monitoring area to generate a production quality correlation interval sequence. S4: Based on the production quality correlation interval sequence and key quality monitoring area, extract the process monitoring segments with overlapping locations, summarize the corresponding surface image features and physical performance measurement contents one by one, combine the monitoring contents at the same location into basic evaluation items, and generate a quality assessment candidate feature set. S5: Call the visual description content and measured index content from the candidate feature set for quality assessment, overlay and compress multiple visual feature descriptions into a single quality change line at the same process position, attach the measured index content to the quality change line at the corresponding position, rearrange all quality segments according to the production sequence, and generate a comprehensive quality assessment report for tarpaulin.

[0021] The defect distribution feature map set includes defect pixel coordinate distribution, image block sequence, and defect type index. The key quality monitoring area band includes continuous abnormal area identification, evaluation priority marker, and process matching sequence. The production quality correlation interval sequence includes process parameter fluctuation information, continuous batch number, and physical index correspondence. The quality assessment candidate feature set includes evaluation item index, visual and physical data pairing results, and process overlapping segment list. The tarpaulin quality comprehensive assessment report includes compression feature line structure, bonding performance index layer, and unified process assessment segment.

[0022] Please see Figure 2 The steps to obtain S1 are as follows: S101: Acquire surface image records of plastic tarpaulin formed during the operation of the collection equipment on the production line, and locate the pixel grayscale data contained in each frame of the image in sequence according to the camera trigger scanning order. According to the preset detection area mask configuration parameter set, the image frame is cropped within the effective detection range, and split into multiple image block segments in the continuous acquisition order. The surface position index data of each segment is numbered to obtain the image position cropping sequence. To acquire surface images of the plastic tarpaulin formed during the operation of the data acquisition equipment on the production line, each frame of the surface image needs to be acquired using a line scan camera or a high-frequency industrial camera. The acquisition sequence and pixel array information must be recorded in real time. The acquisition sequence can be set by sequentially incrementing the frame number from 1 according to the material feed pulse signal. During the acquisition process, a set of mask configuration parameters for the detection area must be set. This parameter set needs to preset the effective detection boundary values ​​for the image, including the lateral... Vertical The range should be set in conjunction with the width of the tarpaulin, such as setting: Based on this mask configuration, the coordinates of the pixels in each frame are determined, where the following condition is satisfied for any pixel: If the selected area is a valid pixel region, the corresponding effective area pixels are retained; otherwise, background masking is performed. The cropped image frames are saved as an original segment sequence in frame number order. To ensure detection continuity, a sampling window step size threshold can be set. For example, set it to 512 pixels. If the overlap between the current capture window and the previous window... If the samples are consecutive, they are considered to be merged into a single image block segment; otherwise, they are split into new segments, each assigned a segment number. Simultaneously, the surface position index information corresponding to each segment is calculated, and the start and end pixel coordinates of each segment in the width direction are recorded as index boundaries. For example, if a segment... middle: Using this range as the boundary index of the cropping result, the positional data of each segment is summarized and saved as an image positional cropping sequence: ,in Corresponding fragments The effective detection range.

[0023] S102: Based on the surface index range of each image block segment in the image position cropping sequence, extract the feature response value of each pixel in the segment, identify the feature dense point group located in the local area, perform response sequence update operation on each group of pixel coordinates by comparing the horizontal and vertical offset relationship, and output the feature point position structure. Each in the image location cropping sequence The surface index range of the corresponding segment is used to read the pixel grayscale matrix it contains, and the feature response value of each pixel is obtained by using a feature map extraction method of a deep convolutional neural network. A sliding window approach is used to extract clusters of dense feature points within a local region of the feature map. For example, the sliding window size is set to... Pixels, the response value within the statistics window is greater than the preset activation threshold. The number of points, if the count is... If the coordinates are not specified, it is determined to be a cluster of feature points (i.e., a potential defect candidate region). For all feature points in the cluster, offset calculations are performed based on their coordinates. For any two feature points... and Calculate its lateral offset separately: Vertical offset: .like and ,in The preset texture continuous step value is considered as adjacent feature points within the same defect connected region. By sequentially comparing the bias relationship between the current point and its neighbors and establishing a chain index, a response sequence update operation is performed to reorder the feature points according to the material feeding direction. The final output feature point order structure contains multiple connected feature point sets, each set representing a spatially logical defect depiction sequence within a local image patch.

[0024] S103: The coordinate sequence of each image block segment in the feature point positional structure is called in the order of its acquisition timestamp to perform a time-series connection operation, and the feature sequences corresponding to multiple segments are connected to the continuous detection channel in chronological order to obtain the defect distribution feature map set; For the coordinate sequence of each segment in the feature point positional structure, a temporal connection operation is performed according to their corresponding acquisition timestamps. First, the synchronization timestamp attribute of each feature point is extracted, and the sequence is sorted in ascending order according to the time value. A threshold for judging temporal continuity is set. Milliseconds, for example, set to 20ms. If the time difference between two adjacent image patches is less than or equal to... If the image is continuously imaged during the material handling process, then a global feature chain is formed by connecting adjacent feature sequences. Continuous defect coordinate paths are extracted from discrete image blocks, and all paths are stitched together along the time axis to form a feature map covering the entire roll of tarpaulin. For example, if the sequence... , , Connect to the continuous detection channel in chronological order: During operation, the geometric continuity of defect features within each segment must be maintained, and the defect density parameters of each segment must be recorded. ,in The number of feature points, Let be the area of ​​the block. If... If the defect is identified as a key defect description segment, the final output defect distribution feature map set consists of multiple defect feature distribution maps with spatiotemporal continuity.

[0025] Please see Figure 3 The steps to obtain S2 are as follows: S201: Based on adjacent segment data in the defect distribution feature map set, compare the corresponding feature point information in each segment in sequence, retrieve the feature response value difference in the horizontal and vertical directions, and perform item-by-item difference confirmation operation with the preset quality offset comparison range. Screen out all segments whose differences exceed the quality discrimination threshold and obtain the abnormal feature index column. Based on adjacent segment data in the defect distribution feature map set, the feature response vectors within each segment need to be extracted sequentially in ascending order of segment number. For each group of adjacent segments, for example... and The average response value of the feature points is extracted and a comparison array is constructed. The difference in feature response values ​​is calculated for each corresponding region, i.e.: Then, the variance and peak value of all response differences in this group of points are statistically analyzed and compared one-to-one with a preset quality offset control range parameter. For example, if the preset control range is that the response fluctuation does not exceed... Then, for each group of adjacent image blocks, determine its... The current segment is recorded as a candidate segment for quality anomalies. Then, all difference judgment results are statistically summarized, and segments exceeding the quality discrimination threshold are filtered out. The discrimination threshold is set according to the tarpaulin grade; for example, 0.15 for Grade 1 and 0.3 for Grade 2. When the proportion of pixels judged to be exceeding the limit in a segment exceeds... When an image block contains 4096 pixel response values, 1500 of which have response differences exceeding the limit, the proportion is calculated as: 1500 / 4096≈0.36. If it is greater than the baseline value of 0.35, then the segment number is filtered and added to the abnormal feature index column.

[0026] S202: Call the fragment number in the abnormal feature index column, extract the data frame of adjacent number according to the production direction, identify all fragment combinations with consecutive numbers and located in the same detection area, determine whether the number of consecutive occurrences in the sequence exceeds the regional connectivity benchmark value, output the number set for fragment combinations that meet the sequence continuity rule, and obtain the abnormal fragment chain set list. The fragment number in the anomaly feature index column is used to scan the data frame corresponding to each number in the original production sequence, extracting the lateral detection position of each fragment on the tarpaulin surface. For example, the fragment number... and Corresponding to the detection areas: and If both are located in the same longitude region and their numbers appear consecutively, they are considered a potential group of consecutive defects. The regional connectivity baseline is set to 4, meaning the number of consecutive anomalous blocks must satisfy the following: When consecutive number combinations, for example When the number of consecutive segments is 4, it is determined that there is a severe strip-shaped or regional defect, and the output is an abnormal segment chain. If the number of consecutive segments is only 2 or 3, it is determined to be isolated point noise and is discarded. This operation ensures that each combination is continuous in both time and space, thereby obtaining segment combinations that meet the rules and forming a list of abnormal segment chains.

[0027] S203: Based on the number data in the abnormal fragment chain set list, perform physical position matching operation on the area corresponding to each chain in the tarpaulin surface texture data, and group the fragment chains with continuous positions according to the surface warp and weft structure connection relationship, and assign key attribute identification information to obtain the key quality monitoring area zone. Based on the serial numbers in the abnormal fragment chain set list, the production material feed coordinates corresponding to each serial number are retrieved one by one. The tarpaulin surface texture data is then used to determine whether adjacent fragments in the chain are located within the same warp and weft weaving texture unit. The matching method involves overlaying the fragment index boundary with the weaving width layer and determining whether the coordinate offset between adjacent fragments in the warp or weft direction is less than the texture step value. If the texture continuity is satisfied, then consecutively positioned fragment chains will be merged and assigned key attribute identifier information. The identifier field is set as follows: This is used to identify areas that fall under the key quality monitoring category. The final result is a key quality monitoring area band, which consists of multiple regions with interconnected textures, continuous positions, and belonging to an abnormal fragment chain. Within each band, the material feeding numbers are continuous and the texture logic is consistent.

[0028] Please see Figure 4 The steps to obtain S3 are as follows: S301: Acquire the physical performance monitoring records generated during the operation of the plastic tarpaulin production equipment, and arrange all records in chronological order into a process data frame sequence. Extract parameter fluctuation information for each record in the sequence and perform trend judgment operation on the direction change of the fluctuation vector. By comparing the stability of the slope sign between adjacent fluctuation vectors, confirm the process segments with consistent trends, and add corresponding process position identification codes to these segments to obtain the trend identification process segment column. Acquire physical property monitoring records generated during the operation of the plastic tarpaulin production equipment, including extrusion temperature, tension values, and composite pressure. Extract these records chronologically and construct a process data frame sequence. Organize the parameter information in each record into a fluctuation vector form. For the data in the sequence, extract the fluctuation vector components. The fluctuation vector of two adjacent records and Calculate its slope respectively The criterion for judging a consistent trend is to compare the signs of adjacent slopes. If multiple consecutive record points satisfy the condition... If the process parameter is in a monotonically increasing or decreasing trend with a stable direction, it is considered to be in a continuous trend. The start and end timestamps of such continuous segments are extracted, and a location identifier code is generated based on the equipment's position in the process (e.g., coating process, heat sealing process). The code format is defined as process ID + time period number, such as "Heat-03" for the third segment of the heat sealing process. This ultimately forms a trend-identified process segment list.

[0029] S302: Call the process position identifier code in the trend identifier process segment column, sequentially retrieve the code data of each process segment, extract the continuously numbered process segments according to the arrangement order of the process segments in the production line, confirm the set of process segments whose number change range is within the process continuity reference value through the sequence continuity judgment operation, and output the process position sequence of these process segments to obtain the continuous production process space column. The system retrieves the identifier codes from the trend indicator process segment column and searches for each process segment in production line sequence. For adjacent process segments, such as "Coat-05" and "Coat-06", it parses their corresponding material feed length coordinates and determines whether they belong to a continuous interval in the production line spatial index. The process continuity baseline is set to a position index change of no more than 1 meter unit. The determination method is based on whether the process segment... and Satisfy in meter coordinate index If the number of continuous process segments reaches the minimum continuous standard value of 3, it is considered a valid process space sequence. During operation, a sliding window check is performed on all identified segments to ensure the consistency of the production rhythm. If the conditions are met, they are included in the continuous production process space column and saved in both process ID and meter coordinates.

[0030] S303: Based on the process position sequence in the continuous production process space column, and the process position data of each segment in the critical quality monitoring area, perform a one-to-one comparison operation on the coordinates of the two under the same production line index system, extract the position segment of the continuous process space sequence in the critical quality monitoring area from the correspondence, and arrange the output sequence according to the production order to obtain the production quality related interval sequence. Based on the process position sequence in the continuous production process spatial column, the meter coordinates of each segment within the critical quality monitoring zone are determined. The start and end coordinates of each process segment in a unified production line index coordinate system are extracted, and each coordinate segment is compared one-to-one within the critical quality monitoring zone. The point-by-point comparison method involves checking the coordinates of each point in the process segment to see if a visually abnormal segment containing that point exists within the monitoring zone. If it exists, the corresponding visual feature number is extracted and associated with process parameter fluctuation information, forming a mapping list from production trajectory to appearance defects. When multiple process segments consecutively correspond to the same monitoring zone, they are sorted according to the material flow direction, and the corresponding segment numbers are extracted to form a stable association sequence. The output sequence is "Process Segment A → Defect Zone B," obtaining the production quality association interval sequence.

[0031] Please see Figure 5 The steps to obtain S4 are as follows: S401: Based on the production quality correlation interval sequence and the key quality monitoring area zone, call the process position index information of each segment in the production quality correlation interval, compare its position number with the position number in the key quality monitoring area zone under the production line index system, perform a screening operation on the data frames with the same coordinate number, extract the segment content with overlapping process positions, and obtain a list of overlapping process segments. Based on the production quality correlation interval sequence and the key quality monitoring area band, the spatial location index parameters of both are extracted. The production quality correlation segments are recorded in structure form. The monitored area is a segment. For each and Perform coordinate comparison to check the meter coordinates between the two on the production line. Does there exist an intersection on the above? If it satisfies... If the two types of data overlap in process position, it is determined that they are identical. For segments that meet the overlap condition, screening is performed, and their complete visual feature vectors and process parameter vectors are extracted and the corresponding number pairs are saved. If the process positions are completely aligned, a list of overlapping process segments is constructed for subsequent multi-source data fusion analysis.

[0032] S402: Based on the set of numbers in the list of overlapping process segments, extract the visual description content associated with the corresponding segment in the surface image feature set and the measured index content in the physical index dataset in sequence. Aggregate the two types of monitoring content with the same process index in numerical order and compress them into an evaluation item structure to obtain the visual dual-mode evaluation item set. Based on the set of numbers in the list of overlapping process segments, for each pair of matching numbers, the corresponding visual description content (such as the defect geometric feature matrix) and measured physical indicators (such as the measured thickness and tensile force at the corresponding location) are retrieved from the database. These two types of heterogeneous data are merged into a unified entry structure, arranged in order of process length during aggregation. To improve evaluation efficiency, data compression is performed on the merged structure. The visual feature sampling rate is set. The physical index data undergoes mean-compression processing. For example, the tension fluctuations of the original 100 sampling points are compressed into 10 key feature points. Each set of entries uses the process number as the primary key and is recorded as an entry item: Process ID → Compressed Image Features → Physical Measured Value. This process is executed cyclically to generate a set of visual dual-mode evaluation entries.

[0033] S403: Call the process number information in the visual dual-mode evaluation item set, arrange all evaluation item structures according to the production flow index order, organize the item sequence according to the number change trend, output the evaluation item dataset with continuous process number characteristics, and obtain the quality evaluation candidate feature set. The process number information from the visual dual-mode evaluation item set is retrieved, and the items are sorted according to the production flow direction (i.e., the direction of meter increase). The sorting is based on the displacement increment implicit in the process number. By parsing each number, the continuity of the processes is confirmed. If the numbers are continuous, then... If the items belong to the same production batch, they are identified as a sequence of items with continuous process characteristics. An ordered dataset of items is then constructed, and sequences that meet the criteria are combined into a candidate feature set. For example, if a group of items is numbered {Batch01-101, Batch01-102, Batch01-103}, and the process length in meters increases sequentially, then the output is a candidate feature set for quality assessment with temporal logic.

[0034] Please see Figure 6 The steps to obtain S5 are as follows: S501: Call the visual description content in the candidate feature set for quality assessment, extract multidimensional response values ​​for the feature line data of all items under the same process number in numerical order, and sequentially stack and process the quality evolution paths of all co-position line segments along the production direction, compress to form an assessment unit structure, and obtain a compressed quality evolution line set. For each process number in the entry, extract the multidimensional response value sequence from its visual feature line data. The data structure is as follows: Visual feature items with the same process position are arranged in ascending order by process number and then stacked vertically. The stacking method involves extracting the peaks and troughs of the response of each feature under the current number and calculating its quality evolution path in the production direction. By repeatedly stacking, multiple scattered defect distribution lines are compressed into a unified quality evolution path. This path preserves the complete trajectory of the defect from its inception to its evolution and is represented by a simplified set of evolution trend points. For example, three related scratch feature lines are compressed into a core evolution line, and finally, a compressed quality evolution line set is constructed according to the process sequence.

[0035] S502: Based on the path location index in the compressed quality evolution line set, call the measured index content under the corresponding number, locate the index change limit value on both sides of each path, and perform close-fitting processing on each index boundary and quality evolution path line, output the combined item structure composed of index boundary and quality line, and obtain the performance fit evaluation item list. Based on the path location index in the compression quality evolution line set, the corresponding physical measured indicators (such as peel strength and folding endurance) are called. On both sides of the quality evolution path, the fluctuation range of the measured indicators is used to locate the indicator variation limits. The bonding operation maps the values ​​of the physical indicators to the coordinate system of the visual evolution line, calculating the logical distance between the path point and the indicator limit. If the fluctuation trend of the physical indicators is highly synchronized with the evolution path of the visual defect in the production direction (i.e., correlation coefficient...), then... If the boundary line of the indicator is to be closely aligned with the quality evolution line, then a combined item structure containing process number, visual evolution path, left physical performance limit, and right physical performance limit is constructed and recorded as a performance alignment evaluation item column.

[0036] S503: Call the number sequence information in the performance fit evaluation item column, extract all combined items according to the continuous process direction, construct a continuous quality evaluation structure according to the arrangement order of the numbers in the production axis, integrate all combined items to form a continuous sequence, and obtain a comprehensive quality evaluation report of tarpaulin. Perform a final sorting of the numbers in the fit evaluation item list to ensure they conform to the axial physical sequence of tarpaulin production. Check that the numbering is continuous and without gaps; if the coordinate changes meet the requirements... If the conditions are met, it is determined to be a continuous quality assessment structure. All combined items are extracted and organized into a structured sequence. Each item in the sequence includes the defect distribution pattern, the corresponding physical performance level, and a process stability evaluation. Finally, all items are aggregated into an assessment report covering the entire production cycle, which is the comprehensive tarpaulin quality assessment report.

[0037] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A comprehensive quality evaluation method for plastic tarpaulins based on deep learning, characterized in that, Includes the following steps: S1: Obtain image records of appearance defects formed on the plastic tarpaulin production line, divide them into continuous image blocks according to a fixed sampling window, extract the defect feature positions in each image block and rearrange them according to row and column coordinates to form block-level quality description results, and connect all block-level quality description results in the production material flow sequence to generate a defect distribution feature map set. S2: Based on the defect distribution feature map set, compare the feature response differences of adjacent image blocks, filter out image groups whose response values ​​exceed the quality discrimination threshold, organize them into continuous regions according to the physical connection relationship of the tarpaulin surface texture, mark the continuous regions with abnormal associations, and generate key quality monitoring area zones. S3: Acquire the physical performance monitoring records collected by sensors during the production process and connect the process trajectory in series according to the production batch time. Extract the production segments with consistent fluctuation trends of process parameters and overlay the corresponding production process identifiers. Integrate the process position of the production segment with the spatial position of the key quality monitoring area to generate a production quality correlation interval sequence. S4: Based on the production quality correlation interval sequence and key quality monitoring area band, extract the process monitoring segments with overlapping positions, summarize the corresponding surface image features and physical performance measurement content one by one, combine the monitoring content at the same position into evaluation basic items, and generate a quality assessment candidate feature set.

2. The method for comprehensive quality evaluation of plastic tarpaulins based on deep learning according to claim 1, characterized in that: The defect distribution feature map set includes defect pixel coordinate distribution, image block sequence, and defect type index. The key quality monitoring area band includes continuous abnormal area identification, evaluation priority marker, and process matching sequence. The production quality correlation interval sequence includes process parameter fluctuation information, continuous batch number, and physical index correspondence. The quality assessment candidate feature set includes evaluation item index, visual and physical data pairing results, and process overlapping segment list.

3. The method for comprehensive quality evaluation of plastic tarpaulins based on deep learning according to claim 1, characterized in that, The steps for obtaining S1 are as follows: S101: Acquire surface image records of plastic tarpaulin formed during the operation of the collection equipment on the production line, and locate the pixel grayscale data contained in each frame of the image in sequence according to the camera trigger scanning order. According to the preset detection area mask configuration parameter set, the image frame is cropped within the effective detection range, and split into multiple image block segments in the continuous acquisition order. The surface position index data of each segment is numbered to obtain the image position cropping sequence. S102: Based on the surface index range of each image block segment in the image position cropping sequence, extract the feature response value of each pixel within the segment. It identifies clusters of dense feature points located within a local area, performs response sequence update operations on each group of pixel coordinates by comparing the horizontal and vertical offset relationships, and outputs the feature point positional structure. S103: The coordinate sequence of each image block segment in the feature point positional structure is called to perform a time-series connection operation according to the acquisition timestamp order, and the feature sequences corresponding to multiple segments are connected to the continuous detection channel in chronological order to obtain a defect distribution feature map set.

4. The method for comprehensive quality evaluation of plastic tarpaulins based on deep learning according to claim 1, characterized in that, The steps for obtaining S2 are as follows: S201: Based on the adjacent segment data in the defect distribution feature map set, the corresponding feature point information in each segment is compared in sequence, the difference in feature response values ​​in the horizontal and vertical directions is retrieved, and the difference is confirmed item by item with the preset quality offset comparison range. All segments with differences exceeding the quality discrimination threshold are screened out to obtain the abnormal feature index column. S202: Call the fragment number in the abnormal feature index column, extract the data frame of adjacent number according to the production direction, identify all fragment combinations with consecutive numbers and located in the same detection area, determine whether the number of consecutive occurrences in the sequence exceeds the regional connectivity benchmark value, output the number set for fragment combinations that meet the sequence continuity rule, and obtain the abnormal fragment chain set list. S203: Based on the numbered data in the abnormal fragment chain set list, perform a physical position matching operation on the area corresponding to each chain in the tarpaulin surface texture data, and group the continuously connected fragment chains according to the surface warp and weft structure connection relationship, and assign key attribute identification information to obtain the key quality monitoring area zone.

5. The method for comprehensive quality evaluation of plastic tarpaulins based on deep learning according to claim 1, characterized in that, The steps for obtaining S3 are as follows: S301: Acquire the physical performance monitoring records generated during the operation of the plastic tarpaulin production equipment, and arrange all records in chronological order into a process data frame sequence. Extract parameter fluctuation information for each record in the sequence and perform trend judgment operation on the direction change of the fluctuation vector. By comparing the stability of the slope sign between adjacent fluctuation vectors, confirm the process segments with consistent trends, and add corresponding process position identification codes to these segments to obtain the trend identification process segment column. S302: Call the process position identifier code in the trend identifier process segment column, sequentially retrieve the code data of each process segment, extract the continuously numbered process segments according to the arrangement order of the process segments in the production line, confirm the set of process segments whose number change range is within the process continuity reference value range through the sequence continuity judgment operation, and output the process position sequence of these process segments to obtain the continuous production process space column. S303: Based on the process position sequence in the continuous production process space column, and the process position data of each segment in the critical quality monitoring area, perform a one-to-one comparison operation on the coordinates of the two under the same production line index system, extract the position segment of the continuous process space sequence in the critical quality monitoring area from the correspondence, and arrange the output sequence according to the production order to obtain the production quality related interval sequence.

6. The method for comprehensive quality evaluation of plastic tarpaulins based on deep learning according to claim 1, characterized in that, The steps for obtaining S4 are as follows: S401: Based on the production quality correlation interval sequence and the key quality monitoring area, call the process position index information of each segment in the production quality correlation interval, compare its position number with the position number in the key quality monitoring area under the production line index system, perform a screening operation on the data frames with the same coordinate number, extract the segment content with overlapping process positions, and obtain a list of overlapping process segments. S402: Based on the set of numbers in the list of overlapping process segments, extract the visual description content associated with the corresponding segment in the surface image feature set and the measured index content in the physical index dataset in sequence. Aggregate the two types of monitoring content with the same process index in numerical order and compress them into an evaluation item structure to obtain the visual dual-mode evaluation item set. S403: Call the process number information in the visual dual-mode evaluation item set, arrange all evaluation item structures according to the production flow index order, organize the item sequence according to the number change trend, output the evaluation item dataset with continuous process number characteristics, and obtain the quality evaluation candidate feature set.

7. The method for comprehensive quality evaluation of plastic tarpaulins based on deep learning according to claim 1, characterized in that, The method further includes: S5: Call the visual description content and measured index content in the quality assessment candidate feature set, superimpose and compress multiple visual feature descriptions into a single quality change line at the same process position, attach the measured index content to the quality change line at the corresponding position, rearrange all quality segments according to the production sequence, and generate a comprehensive tarpaulin quality assessment report. The comprehensive quality assessment report for the tarpaulin includes the compression feature line structure, the bonding performance index layer, and the unified process assessment segment.

8. The method for comprehensive quality evaluation of plastic tarpaulins based on deep learning according to claim 7, characterized in that, The steps for obtaining S5 are as follows: S501: Call the visual description content in the quality assessment candidate feature set, extract multidimensional response values ​​for the feature line data of all items under the same process number in numerical order, and sequentially stack and process the quality evolution paths of all co-position line segments along the production direction, compress to form an assessment unit structure, and obtain a compressed quality evolution line set. S502: Based on the path location index in the compressed quality evolution line set, call the measured index content under the corresponding number, locate the index change limit value on both sides of each path, and perform close-fitting processing on each index boundary and quality evolution path line, output the combined item structure composed of index boundary and quality line, and obtain the performance fit evaluation item list. S503: Call the number sequence information in the performance fit evaluation item column, extract all combined items according to the continuous process direction, construct a continuous quality evaluation structure according to the arrangement order of the numbers in the production axis, integrate all combined items to form a continuous sequence, and obtain a comprehensive quality evaluation report of tarpaulin.

Images