Pepper segment processing production line internet of things monitoring system and cloud platform

Abstract

The present application relates to the technical field of pepper deep processing, in particular to a pepper segment processing production line internet of things monitoring system and cloud platform, the system contains a sensor monitoring unit, a dynamic weight distribution unit, a slime interference elimination unit, a displacement compensation algorithm unit and a control linkage unit, the sensor monitoring unit in the present application collects the moisture value, position coordinates and probe contact signal of the pepper segment in the dehydration bin grid in real time, the dynamic weight distribution unit identifies the large and small pepper segment clusters and assigns corresponding weights to generate the grid moisture true value, the slime interference elimination unit marks the slime attachment and compensates for the lag bias, the displacement compensation algorithm unit establishes the displacement probability cloud chart to realize the probe displacement compensation, and the control linkage unit regulates the hot air parameters and outputs the mold risk signal, solves the problems of inaccurate moisture monitoring and difficult interference elimination, and improves the processing quality.

Description

Technical Field

[0001] This invention relates to the field of chili deep processing technology, specifically to an Internet of Things (IoT) monitoring system and cloud platform for chili segment processing production lines. Background Technology

[0002] Deep processing of chili peppers is an important technology. In the context of the increasing demand for large-scale and standardized chili pepper processing industry, this technology is a key support to break through the limitations of traditional manual monitoring. It can not only replace the lagging mode of manual sampling and testing of moisture at regular intervals, but also capture subtle changes in the dehydration process through real-time data. It can also avoid the problem of chili pepper segments being too dry or too wet due to human judgment errors. Furthermore, it can improve the automation level of the production line and the consistency of product quality by intelligently adjusting processing parameters.

[0003] Existing monitoring technologies for chili processing lines face core problems in practical applications, including low accuracy in moisture monitoring, difficulty in eliminating interference factors, and an inability to adapt to dynamic changes in chili segments. These issues lead to unstable processing quality and a high risk of mold growth. The key reason for this problem lies in the failure to consider the impact of chili segment morphology differences on evaporation rates. Current systems use uniform sensor weights to collect moisture data. However, large chili segments evaporate slowly due to their size, while small chili segments evaporate quickly. This uniform weighting causes the moisture calculation results to deviate from the true value, failing to reflect the actual dehydration status of different size clusters. Furthermore, the probes are susceptible to mucus interference. During chili segment dehydration, sticky substances are released, adhering to the surface of the moisture probe and forming a mucus film. This causes a lag in the probe contact pressure signal and a delay in moisture data transmission. The system fails to promptly mark the mucus adhesion and compensate for the deviation, further exacerbating data distortion. For the dynamic displacement of chili pepper segments, the hot air flow in the dehydration chamber causes the segments to move within the grid. Existing probes collect data at fixed positions. When the chili pepper segments deviate from the probe positions, data loss or invalid data collection occurs, and there is no compensation mechanism, resulting in incomplete spatial data coverage. This defect leads to inaccurate moisture data due to morphological differences. The system adjusts the hot air parameters based on erroneous data. Large chili pepper segments are prone to being over-wet, while small ones are prone to being over-dry. Mucus interference causes data lag, missing the optimal adjustment time. Over-wet areas remain in high humidity, and displacement leads to data loss, making it impossible to detect local over-wetness or over-dryness in time. The combination of these three factors ultimately leads to serious problems. Over-wet areas breed mold due to the high humidity environment, causing batches of products to be scrapped, increasing production costs. Over-dry areas cause chili pepper segments to become brittle and break, reducing product quality. To solve this problem, we provide an IoT monitoring system and cloud platform for the chili pepper segment processing production line. Summary of the Invention

[0004] The purpose of this invention is to provide an Internet of Things (IoT) monitoring system and cloud platform for chili segment processing production lines to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, an IoT monitoring system for chili segment processing production lines is provided, including:

[0006] The sensor monitoring unit is used to collect the moisture value, spatial coordinates, and probe contact status signals of the chili pepper segments in each grid of the dehydration chamber in real time.

[0007] The dynamic weight allocation unit automatically identifies the spatial partitions of large-sized chili pepper clusters and small-sized chili pepper clusters based on the morphological distribution characteristics of chili pepper segments within the grid obtained by machine vision. It constructs a dynamic weight model to assign incremental and decremental weight coefficients to the moisture data of large-sized and small-sized chili pepper clusters, respectively. The unit generates the true moisture value of the grid by superimposing the spatial partition weights, thus eliminating the error caused by uneven evaporation rate due to morphological differences.

[0008] The mucus interference elimination unit marks mucus adhesion events on the probe surface in real time based on the contact pressure change rate and contact time threshold in the probe contact status signal, triggers the cleaning compensation mechanism, and inversely superimposes the hysteresis deviation caused by the mucus film in the raw moisture data stream, and activates the self-cleaning command to the sensor monitoring unit, outputting the real-time moisture value after mucus interference correction.

[0009] The displacement compensation algorithm unit establishes a displacement probability cloud map of the chili pepper segment based on the temporal displacement trajectory of the chili pepper segment's location coordinates, and synchronously associates the spatial confidence of each probe. When a probe falls into the low confidence zone of the displacement probability cloud map, the probe data is automatically removed and the adjacent probes are activated to fill in the gaps and maintain the spatial coverage integrity of the multi-point data.

[0010] The control linkage unit processes the large-sized chili pepper clusters and the small-sized chili pepper clusters respectively based on the actual value of grid moisture and the spatial partition weight coefficient, and analyzes the slope of the change of hysteresis deviation and the decay rate of spatial confidence in real time. When the superposition value of the two exceeds the tolerance, a mold growth risk signal is output.

[0011] The second objective of this invention is to provide a cloud platform for implementing an Internet of Things (IoT) monitoring system for a chili segment processing production line, comprising any one of the above-mentioned features. The platform is characterized by including: an edge access module, a parallel computing module, a strategy generation module, and a device reverse control module, wherein:

[0012] The edge access module collects and compresses data from the dehydration chamber in real time and uploads it. The parallel computing module runs the sensor monitoring unit, dynamic weight allocation unit, mucus interference elimination unit, displacement compensation algorithm unit, and control linkage unit. The strategy generation module integrates the calculation results to generate control instructions and risk warnings. The equipment reverse control module sends instructions to the dehydration chamber and monitors the execution status.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0014] This invention automatically identifies clusters of large and small chili pepper segments using a dynamic weight allocation unit and assigns increasing and decreasing weights to generate accurate grid moisture values. A mucus interference elimination unit marks mucus adhesion based on the probe contact pressure change rate and contact time, and inversely superimposes the hysteresis deviation while activating self-cleaning. A displacement compensation algorithm unit establishes a displacement probability cloud map based on the chili pepper segment displacement trajectory and activates neighboring probes to fill in when the probe falls into a low-confidence zone. A control linkage unit adjusts hot air parameters according to the accurate grid moisture values ​​and analyzes the slope of the deviation change and the confidence decay rate to output a mold risk signal. This achieves precise moisture monitoring, effective elimination of mucus interference, complete data spatial coverage, stable processing quality, and reduced mold risk. It effectively solves the problems of low moisture monitoring accuracy, difficulty in eliminating interference factors, and inability to adapt to dynamic changes in chili pepper segments, leading to unstable processing quality and high mold risk in chili pepper processing production lines. Attached Figure Description

[0015] Figure 1 This is an overall block diagram of the present invention;

[0016] Figure 2 This is the overall flowchart of the present invention.

[0017] The meanings of the labels in the diagram are as follows:

[0018] 1. Sensor monitoring unit; 2. Dynamic weight allocation unit; 3. Mucus interference elimination unit; 4. Displacement compensation algorithm unit; 5. Control linkage unit. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, 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.

[0020] This invention provides an IoT monitoring system for a chili pepper processing production line. Please refer to [link / reference]. Figure 1 As shown, it includes:

[0021] Sensor monitoring unit 1 is used to collect the moisture value, spatial coordinates and probe contact status signals of the chili segments in each grid of the dehydration chamber in real time;

[0022] The dynamic weight allocation unit 2 automatically identifies the spatial partitions of large-sized chili segment clusters and small-sized chili segment clusters based on the morphological distribution characteristics of chili segments within the grid obtained by machine vision. It constructs a dynamic weight model to assign incremental and decremental weight coefficients to the moisture data of large-sized and small-sized chili segment clusters, respectively. The unit generates the true value of grid moisture by superimposing the spatial partition weights, thus eliminating the error caused by uneven evaporation rate due to morphological differences.

[0023] The mucus interference elimination unit 3 marks mucus adhesion events on the probe surface in real time based on the contact pressure change rate and contact time threshold in the probe contact state signal, triggers the cleaning compensation mechanism, and superimposes the hysteresis deviation caused by the mucus film in the original moisture data stream, and activates the self-cleaning command to the sensor monitoring unit 1, outputting the real-time moisture value after mucus interference correction.

[0024] The displacement compensation algorithm unit 4 establishes a probability cloud map of chili segment displacement based on the temporal displacement trajectory of the chili segment position coordinates, and synchronously associates the spatial confidence of each probe. When a probe falls into the low confidence area of ​​the displacement probability cloud map, the probe data is automatically removed and the adjacent probes are activated to fill in the gaps and maintain the spatial coverage integrity of multi-point data.

[0025] The control linkage unit 5 processes the large-sized chili pepper clusters and the small-sized chili pepper clusters respectively based on the actual value of grid moisture and the spatial partition weight coefficient, and analyzes the slope of the change of hysteresis deviation and the decay rate of spatial confidence in real time. When the superposition value of the two exceeds the tolerance, a mold growth risk signal is output.

[0026] In the dynamic weight allocation unit 2, spatial partitioning of large-sized and small-sized chili pepper segment clusters is automatically identified. Specifically, the machine vision module synchronously collects multi-angle surface texture images and 3D point cloud data of chili pepper segments within the grid. A composite feature extraction algorithm is used to segment the chili pepper segment outlines at multiple scales, and the chili pepper segments are separated from the background area based on the HSV color space. Then, edge curvature analysis is used to perform topological segmentation of the adhesion area of ​​adjacent chili pepper segments. Finally, the minimum circumscribed cube volume of each chili pepper segment is calculated by combining the Z-axis coordinate data obtained by the depth camera. When the volume value is greater than the preset size threshold, it is classified as a large-sized chili pepper segment cluster, and when it is less than the threshold, it is classified as a small-sized chili pepper segment cluster. A vector partition map containing the centroid coordinates of the two types of clusters is generated in real time.

[0027] Further explanation is needed regarding the specific implementation method of the dynamic weight allocation unit in identifying the spatial partitioning of large and small chili pepper segment clusters. After the sensor monitoring unit 1 completes the collection of the moisture value and location coordinates of the chili pepper segments in each grid of the dehydration chamber, the dynamic weight allocation unit 2 needs to first identify the spatial partitioning of large and small chili pepper segment clusters. If the spatial distribution of the two types of clusters cannot be distinguished, the subsequent weight allocation will have errors due to the uneven evaporation rate caused by the morphological differences. The specific implementation method for identification is as follows:

[0028] The machine vision module synchronously acquires multi-angle surface texture images and 3D point cloud data of chili pepper segments within a grid. These two types of data are the foundation for all subsequent processing, and the acquisition process must ensure spatial correspondence. The machine vision module consists of three high-definition industrial cameras and one depth camera, which acquire multi-angle surface texture images and 3D point cloud data once each. This data records the spatial position of each chili pepper segment in the form of 3D coordinates (X, Y, Z), where X and Y correspond to the planar coordinates of the dehydration chamber grid, and Z corresponds to the depth of the chili pepper segment from the camera, i.e., the vertical height of the chili pepper segment. Each set of 3D point cloud data is calibrated with the multi-angle surface texture images acquired at the same time through timestamps and spatial coordinates to ensure that the 2D texture features and 3D position information of a certain chili pepper segment can accurately correspond, laying the data association foundation for the subsequent transition from 2D texture analysis to 3D volume calculation.

[0029] After data acquisition, a composite feature extraction algorithm is first used to segment the chili pepper segments at multiple scales based on multi-angle surface texture images. This algorithm simultaneously extracts edge features (grayscale difference between the chili pepper segment and surrounding areas), texture features (density and depth of surface wrinkles), and color features (distribution of red hues in the chili pepper segments) from the texture image, avoiding segmentation bias caused by single features. The multi-scale segmentation logic is to first coarsely segment and then finely segment. First, a large-scale window of 10×10 pixels is used to traverse the image. Based on edge and color features, cluster candidate regions containing multiple chili pepper segments are initially segmented. For example, if all pixels in a certain region are the characteristic red of chili pepper segments and the edges are continuous, it is considered a candidate region. Then, a small-scale window of 2×2 pixels is used to further subdivide the candidate regions, combining texture features... The system distinguishes the boundaries of different chili pepper segments. For example, two adjacent chili pepper segments may have different texture directions, one with horizontal wrinkles and the other with vertical wrinkles. A small-scale window can capture these subtle differences, thus segmenting the initial outline of a single chili pepper segment within the candidate region. At this point, the obtained outline still includes background areas, such as the metal mesh of the dehydration chamber or the white tray, which need further processing to purify the workpiece area. The chili pepper segments are separated from the background area based on the HSV color space. The HSV color space can more accurately distinguish objects from the background by color and is more resistant to the interference of lighting changes compared to the RGB space. In specific operations, the outline and background images obtained from multi-scale segmentation are first converted from RGB format to HSV format, where H is hue, S is saturation, and V is lightness. Then, a threshold is set according to the color difference between the chili pepper segment and the background, as follows:

[0030] Because chili pepper segments contain capsanthin, their hue (H) value is typically in the range of 0-10° (red) and 160-179° (dark red), their saturation (S) value is greater than 50% to avoid interference from light-colored backgrounds, and their brightness (V) value is greater than 30% to exclude shadow areas. Background areas (metal mesh is silver-gray, with an H value of 180-220° and an S value of <30%, while white trays have an H value of 0-360°, an S value of <20%, and a V value of greater than 80%) fall outside these thresholds. Through threshold filtering, pixels in the contour image that conform to the HSV range of chili pepper segments are retained, while background pixels that do not conform are removed. The final result is a pure contour image containing only chili pepper segments, ensuring that subsequent processing targets only chili pepper segments and avoiding interference from background pixels in segmentation and volume calculation.

[0031] By performing topological segmentation on the adhered regions of adjacent chili pepper segments through edge curvature analysis, the problem of adjacent chili pepper segments adhering still exists in the pure contour image. For example, two small chili pepper segments are tightly attached together, and the contour boundary is connected as one. If not segmented, multiple adhered chili pepper segments will be mistakenly identified as a cluster of large chili pepper segments. The specific segmentation process is as follows:

[0032] The coordinates of edge points for each candidate chili pepper segment contour in the pure contour image are extracted, and the curvature of each edge point is calculated. Curvature is an indicator describing the degree of curvature of a curve. Due to the intersection of the contours of two chili pepper segments, the curvature of the edge points in the adhered region will show a significant abrupt change. The curvature value is calculated by sliding a window of 3 edge points, and points with curvature greater than a preset threshold are selected as potential cutting points. The potential cutting points are topologically verified. If the edge length between two adjacent potential cutting points is less than the minimum diameter of a single chili pepper segment and the line connecting the two points can divide the adhered contour into two independent closed regions, it is determined to be a valid cutting point. The valid cutting points are connected by straight lines to topologically divide the adhered contour into two independent chili pepper segment contours, which ensures the accuracy of subsequent volume calculation.

[0033] After accurately segmenting the outline of a single chili pepper segment, the minimum bounding cube volume of each segment is calculated using Z-axis coordinate data acquired by a depth camera. The minimum bounding cube is the smallest cube that can completely enclose a single chili pepper segment, and its volume directly reflects the size of the segment. The calculation process relies on the established correspondence between the 2D outline and the 3D point cloud. The specific process is as follows:

[0034] Based on the 2D contour coordinates (X,Y) of a single chili pepper segment in a pure contour image, all 3D points belonging to that chili pepper segment are selected from the concurrently acquired 3D point cloud data. Through spatial coordinate matching, the pixels within the 2D contour are mapped to the (X,Y) range of the 3D point cloud. All Z-axis coordinates corresponding to these (X,Y) coordinates are extracted to determine the 3D coordinate range of the chili pepper segment. The X-axis range is the maximum and minimum values ​​of the corresponding 3D point X-coordinate, the Y-axis range is Ymax-Ymin, and the Z-axis range is Zmax-Zmin. The length, width, and height of the minimum bounding cube are calculated, where length = Xmax-Xmin, width = Ymax-Ymin, height = Zmax-Zmin, and volume = length × width × height. The calculated volume values ​​are compared with preset size thresholds to complete the size classification of a single chili pepper segment. Finally, a vector partition map containing the centroid coordinates of the two clusters is generated in real time. This map is the spatial basis for subsequent weight allocation and the spatial distribution of the two clusters needs to be clearly marked. The specific generation process is as follows:

[0035] All chili pepper segments are grouped into clusters based on size. Chili pepper segments of the same size with a spatial distance of less than 5 mm (the typical spacing of chili pepper segments in the dehydration chamber) are grouped into one cluster. For example, if there are 8 large chili pepper segments in a grid, 5 of them are spaced less than 5 mm apart, forming one large chili pepper segment cluster, and the remaining 3 are independent (spaced more than 5 mm apart), forming a total of 4 large chili pepper segment clusters. The centroid coordinates of each cluster are calculated. The centroid coordinates are the average of the three-dimensional coordinates (X, Y, Z) of all chili pepper segments within the cluster. For example, a large chili pepper segment cluster contains 3 chili pepper segments with coordinates (25, 45, ...). Given (4.5), (27,47,4.8), and (26,46,4.6), the centroid coordinates are ((25+27+26) / 3, (45+47+46) / 3, (4.5+4.8+4.6) / 3) = (26,46,4.6). A vector partition map is constructed, using the planar coordinates (X,Y) of the dehydration chamber grid as the reference. Different colored vector points represent clusters of large-sized chili pepper segments: red for large-sized clusters and blue for small-sized clusters. The number of chili pepper segments and average volume of each cluster are labeled next to the vector points. For example, a red vector point might be labeled "large-sized - 5 - 280mm". 3 The blue vector dots are labeled with small dimensions - 3 - 90mm. 3 Furthermore, the map is updated in real time according to timestamps, ensuring that the subsequent dynamic weight model can accurately assign weight coefficients to chili pepper clusters of different sizes in different locations based on the latest spatial partition information, thus eliminating the error of uneven evaporation rate caused by morphological differences.

[0036] The method for constructing the dynamic weight model in dynamic weight allocation unit 2 specifically includes:

[0037] Based on the vector partition map, the proportion of large-sized chili pepper clusters to small-sized chili pepper clusters is statistically analyzed. A cluster size distribution ratio function is established, and the weight coefficient of large-sized chili pepper clusters is set to increase positively correlated with the proportion of their quantity, while the weight coefficient of small-sized chili pepper clusters is set to decrease negatively correlated with their quantity. At the same time, a real-time evaporation rate compensation factor is introduced, and the slope of the weight coefficient is dynamically adjusted according to the hot air temperature gradient in the dehydration chamber to ensure that large-sized chili pepper clusters receive additional weight increase under low evaporation rate conditions.

[0038] Further explanation is needed regarding the specific implementation of the dynamic weight allocation unit in constructing the dynamic weight model. After the dynamic weight allocation unit 2 generates a vector partition map containing the centroid coordinates of large and small chili pepper clusters, a dynamic weight model needs to be constructed based on this map. This ensures that the moisture data weights for clusters of different sizes are adapted to their morphological characteristics and dehydration environment, avoiding misjudgments of moisture values ​​due to the slower evaporation of large chili pepper clusters and the faster evaporation of small chili pepper clusters. The specific implementation method is as follows:

[0039] Based on the vector partition map, statistically analyzing the proportion of large-sized and small-sized chili pepper clusters is a crucial step in determining the basis for weight allocation. Before the statistical analysis, it's essential to define the statistical scope as a single grid within the dehydration chamber. Weights are calculated independently for each grid to avoid interference from cross-grid environmental differences. First, the total number of both types of clusters within the current grid is extracted from the vector partition map. For example, if a grid is marked with 4 large-sized chili pepper clusters and 6 small-sized chili pepper clusters, with each large-sized cluster containing 3 to 5 chili pepper segments and each small-sized cluster containing 2 to 3 segments, then the number of large-sized chili pepper clusters is 4, the number of small-sized chili pepper clusters is 6, and the total number of clusters is 10. Next, the proportion of each type of cluster is calculated. The proportion is the percentage of a particular type of cluster relative to the total number of both types, calculated by dividing the number of a particular type of cluster by the total number of both types. Therefore, the proportion of large-sized chili pepper clusters is equal to... The proportion of small-sized chili segments in the cluster is equal to Furthermore, the statistical process is synchronized with the update frequency of the vector partition map to ensure that the proportion data can reflect the dynamic distribution changes of cluster size within the grid in real time. After the statistics are completed, a cluster size distribution ratio function is established. The core function of this function is to transform the basic statistical value of quantity proportion into a ratio coefficient that can be directly used for weight calculation, avoiding insufficient adaptability caused by the direct linear correspondence between proportion and weight. The specific construction process is as follows:

[0040] The input and output variables of the function are determined. The input is the proportion of large or small chili pepper clusters, ranging from 0 to 100%. The output is the corresponding cluster size ratio coefficient, ranging from 0.5 to 1.5. Historical data is collected for function fitting. The correlation data between the proportion of the same grid and the optimal weight coefficient is retrieved from the past 30 days. The optimal weight coefficient is the coefficient value that minimizes moisture error, verified by experiments. A linear fitting method is used to establish the functional relationship. For example, the proportion function of large chili pepper clusters is set to a ratio coefficient equal to 0.7 + 0.01 × proportion of the number of chili peppers, and the proportion function of small chili pepper clusters is set to a ratio coefficient equal to 1.3 - 0.005 × proportion of the number of chili peppers. The fitted ratio coefficients are then substituted into the historical data. In the calculation of moisture data, if the moisture value error rate decreases from 15% to below 5%, the function is confirmed to be effective; otherwise, the fitting parameters are adjusted until the error reaches the target, ultimately forming a cluster size distribution ratio function that can stably output the ratio coefficient. Based on the ratio function, basic weight coefficients are set for the two types of clusters. The weight coefficient of the large-sized chili pepper cluster increases positively with its quantity proportion, while the weight coefficient of the small-sized chili pepper cluster decreases negatively with its quantity proportion. This is because a higher quantity proportion of large-sized chili pepper clusters indicates more chili pepper segments with slow evaporation within the grid, requiring a higher weight to highlight their impact on the overall grid moisture content. Conversely, a higher quantity proportion of small-sized chili pepper clusters indicates more chili pepper segments with fast evaporation, requiring a lower weight to avoid... To avoid interference from rapidly changing moisture data in the overall judgment, the specific operation involves first setting a baseline value for the weighting coefficient, calibrated based on historical average moisture data. The initial baseline value for both large and small chili pepper clusters is 1.0. Then, the proportional coefficient output by the cluster size distribution proportional function is multiplied by the baseline value to obtain the final basic weighting coefficient. If the proportion of large chili pepper clusters subsequently increases to 50%, the proportional function output coefficient becomes 1.2, and the basic weighting coefficient is increased accordingly to 1.2, ensuring that the weight can be dynamically adjusted according to the proportion. At this point, the basic weighting coefficient only considers the cluster size distribution and does not take into account the real-time evaporation environment inside the dehydration chamber. However, the hot air temperature gradient directly affects the evaporation rate, especially for large chili pepper clusters at low evaporation rates. Moisture content changes more slowly, and without adjusting the weights, moisture data feedback can easily lag. Therefore, a real-time evaporation rate compensation factor needs to be introduced. This factor is a coefficient dynamically calculated based on the hot air temperature gradient within the dehydration chamber, with a value ranging from 0.8 to 1.5. The slope of the base weight coefficient is adjusted, i.e., the rate at which the weight changes with its proportion, so that the weight can adapt to different evaporation rate conditions. In practice, five temperature sensors are evenly distributed within the dehydration chamber grid, located at the four corners and the center of the grid, to collect real-time hot air temperatures. The temperature difference between adjacent sensors is calculated, and the average of all adjacent differences is the hot air temperature gradient of the current grid. A correlation rule is established between the temperature gradient and the compensation factor. When the temperature gradient is greater than 3°C, the compensation factor is set to 0.9. At this point, the evaporation rate is fast, and no additional weight increase is needed. The weight slope should be appropriately reduced. When the temperature gradient is 1-3℃ (medium evaporation rate), the compensation factor is set to 1.0 to maintain the basic slope. When the temperature gradient is less than 1℃ (low evaporation rate, uneven hot air distribution), the compensation factor is set to 1.2, and the weight slope needs to be increased to provide additional support for large-sized chili pepper clusters. The slope of the basic weight coefficient is adjusted using the compensation factor. For example, if the basic weight slope for large-sized chili pepper clusters is 0.01 (the weight increases by 0.01 for every 1% increase in proportion), when the compensation factor is 1.2, the adjusted slope = 0.01 × 1.2 = 0.012. (For every 1% increase in proportion, the weight increases by 0.012). If the proportion of large-sized chili pepper clusters increases from 40% to 50%, the weight coefficient increases from 1.1 (40% × 0.012 + 0.7 = 1.18, recalculated based on the adjusted slope) to 1.3 (50% × 0.012 + 0.7 = 1.3). Compared to the uncompensated 1.2, this represents an additional 0.1 increase in weight, ensuring that the moisture data of large-sized chili pepper clusters is more fully represented under low evaporation rates, avoiding misjudgments of moisture values ​​due to slow evaporation, and ultimately forming a dynamic weight model that considers both size distribution and environmental conditions.

[0041] In dynamic weight allocation unit 2, when generating the true grid moisture value through spatial partition weight superposition, a partition moisture value fusion algorithm is adopted, which specifically includes:

[0042] Extract the probe moisture values ​​corresponding to the centroid coordinates of each cluster in the vector partition map. Weight the high weight coefficient of the large-sized chili pepper cluster with its moisture value and weight the low weight coefficient of the small-sized chili pepper cluster with its moisture value. Then perform spatial interpolation on the two types of weighted values. Combine Kalman filtering to eliminate hot wind disturbance noise and finally output smooth and continuous grid moisture values.

[0043] Further explanation is needed regarding the specific implementation method of generating grid moisture values ​​using the partition moisture value fusion algorithm in the dynamic weight allocation unit. After the dynamic weight allocation unit 2 completes the spatial partitioning identification of large and small chili pepper segment clusters and the construction of the dynamic weight model, it needs to convert the discrete cluster moisture data into continuous grid moisture values ​​using the partition moisture value fusion algorithm. If the moisture data of a single probe is used directly, some areas within the grid will not have data coverage due to the probe placement spacing, and the weight differences of clusters of different sizes are not considered, which can easily lead to moisture value distortion. Therefore, it is necessary to proceed according to the process. The specific implementation method is as follows:

[0044] Extracting the probe moisture values ​​corresponding to the centroid coordinates of each cluster in the vector partition map is fundamental to ensuring accurate correlation between moisture data and cluster size. The three-dimensional coordinates of the centroid of each chili pepper segment cluster of different sizes are marked in the vector partition map. The moisture probes of sensor monitoring unit 1 are evenly arranged according to the dehydration chamber grid. Therefore, it is necessary to determine the probe corresponding to each cluster through coordinate matching. Using the cluster centroid coordinates as a reference, the moisture probe closest to the centroid is retrieved, and the moisture value collected by the probe in real time is used as the reference moisture value of the cluster. If there are two or more probes within 50mm of the centroid of a cluster, the average of the moisture values ​​of these probes is taken as the reference moisture value to ensure that each cluster has a unique and accurate source of moisture data and to avoid errors in subsequent weighted calculations due to probe data deviations.

[0045] Weighted summaries are performed based on cluster size type, allowing different weight coefficients to reflect their varying contributions to the overall moisture content of the grid. For large-sized chili pepper clusters, the high-weight coefficient assigned to them in the dynamic weighting model is retrieved and multiplied by the baseline moisture value of the corresponding cluster to obtain the weighted moisture value for the large-sized chili pepper cluster. For small-sized chili pepper clusters, the corresponding low-weight coefficient is retrieved and multiplied by the baseline moisture value to obtain the weighted moisture value for the small-sized chili pepper cluster. The weighted moisture values ​​of all large-sized chili pepper clusters within the same grid are summed, and the weighted moisture values ​​of all small-sized chili pepper clusters are also summed to obtain the total weighted moisture value for each type of cluster. The logic behind this step is that large-sized chili pepper clusters, due to their slower evaporation, have a greater impact on the overall moisture content, and therefore, the high weight coefficient is used to calculate the weighted moisture value. Weighting coefficients amplify the contribution of moisture data, while small-sized chili pepper clusters are balanced by low weighting coefficients to mitigate moisture fluctuations caused by rapid evaporation, ensuring that the accumulated result reflects the true impact of different-sized clusters on the grid's moisture content. After weighted accumulation of the two types of clusters, the resulting value is still a discrete total weighted moisture value. The dehydration chamber grid requires complete and continuous moisture distribution data to accurately reflect the overall dehydration state. Therefore, spatial interpolation is used to fill the gaps between discrete points. Spatial interpolation is a technique that uses known discrete point moisture values ​​to mathematically estimate the moisture value of unknown regions. Here, inverse distance weighted interpolation is used. This method ensures that unknown points closer to known points are more significantly affected by the moisture value of known points, resulting in a result that better reflects the actual moisture distribution pattern. The specific process is as follows:

[0046] The original grid of the dehydration chamber (e.g., 100×100mm) is divided into a finer grid of 1×1mm, resulting in 10,000 fine grid points. The moisture value for each point needs to be calculated. For each fine grid point, only the weighted moisture values ​​of known clusters within a 20mm radius centered on that point are considered. The influence of clusters outside this radius on the moisture value of that point is negligible, avoiding interpolation bias caused by distant clusters. The inverse of the straight-line distance between the fine grid point and the centroid of each cluster within the range is calculated, and then all inverses are normalized. The weighted moisture value of each cluster is multiplied by its corresponding distance weight, and all products are summed to obtain the interpolated moisture value for that fine grid point. This process is repeated until moisture values ​​are calculated for all fine grid points. At this point, a continuous moisture distribution matrix is ​​formed within the grid, resolving the problem of incomplete coverage of discrete data. However, due to dynamic disturbances in the hot air within the dehydration chamber, instantaneous noise may appear in the interpolated moisture data. Direct output would... This noise affects the judgment of the subsequent control linkage unit 5, so it is necessary to combine it with Kalman filtering to eliminate such noise. Kalman filtering is an algorithm that suppresses noise by establishing state equations and observation equations and fusing predicted and observed values. Here, the moisture value of the interpolated fine grid points is used as the observed value. In specific operation, the moisture value at the next moment is predicted based on the dehydration law of the chili pepper segment. The prediction error is set based on the fluctuation range of historical moisture changes, and the observation error is based on the statistical moisture fluctuation caused by hot air disturbance. The predicted value and the observed value are fused through Kalman gain to obtain the filtered moisture value. After performing Kalman filtering on the moisture value of all fine grid points one by one, the final output is a smooth and continuous true value of grid moisture. This true value not only eliminates the error of uneven evaporation rate caused by morphological differences, but also filters out the noise of hot air disturbance, providing accurate moisture data support for the subsequent control linkage unit 5 to adjust hot air parameters for different size clusters and judge the risk of mold growth.

[0047] The slime interference cancellation unit 3 marks slime adhesion events based on probe contact status signals, specifically including:

[0048] The probe contact pressure change rate is monitored in real time. When the pressure change rate shows an exponential decay characteristic of first rising sharply and then falling slowly, it is determined to be the initial adhesion of mucus. At the same time, the continuous contact time of the probe is detected. If the time exceeds the typical contact cycle under drying conditions, the mucus thickening judgment is triggered. When both characteristics are met at the same time, a mucus adhesion event marker signal is generated.

[0049] Further explanation is needed regarding the specific implementation of the mucus interference elimination unit 3 marking mucus adhesion events based on the probe contact state signal. After the dynamic weight allocation unit 2 outputs smooth and continuous grid moisture values ​​through the partitioned moisture value fusion algorithm, the moisture probe of the sensor monitoring unit 1 needs to continuously contact the chili pepper segment to maintain data acquisition. However, the chili pepper segment will release mucus during dehydration. The polysaccharides contained in this mucus seep out with the moisture, resulting in a viscous substance that easily adheres to the metal detection end of the probe. This mucus will change the contact state between the probe and the chili pepper segment, causing a lag deviation in subsequent moisture acquisition. Therefore, the mucus interference elimination unit 3 needs to accurately mark the mucus adhesion event based on the probe contact state signal. This signal is data collected in real time by the sensor monitoring unit 1, reflecting the physical interaction between the probe and the chili pepper segment during contact. It includes information such as the contact pressure change curve over time and the timestamps of the start and end of the contact. The specific marking process needs to be implemented through dual logic of pressure change feature judgment and contact duration verification to avoid misjudgment caused by a single feature. The specific implementation is as follows:

[0050] Real-time monitoring of the probe contact pressure change rate is crucial. The contact pressure change rate refers to the rate of change of contact pressure over time during the probe's contact with the chili pepper segment. Its characteristics directly reflect the presence of mucus on the probe surface. Under drying conditions, the pressure of the probe contacting the chili pepper segment quickly rises to a stable value, and then the pressure change rate remains relatively stable, close to zero, without significant decay. However, when mucus begins to adhere to the probe surface, its viscosity hinders the rapid and stable contact between the probe and the chili pepper segment. In the initial stage of contact, the mucus rapidly increases the contact area between the probe and the chili pepper segment, causing a sharp increase in contact pressure within a short period. In the later stage of contact, the elastic deformation of the mucus causes the pressure to decrease slowly, exhibiting a quasi-exponential decay characteristic of a sharp initial rise followed by a gradual decrease, i.e., the pressure change rate. The slope rapidly decreases from a high positive value during a steep rise to a low negative value during a gradual decline, and the decay process lasts for more than 0.2 seconds. Under drying conditions, it has stabilized at this point. The mucus interference elimination unit 3 extracts pressure change rate data every 10 milliseconds and determines whether it conforms to the exponential decay characteristic by using a preset slope threshold. If the data extracted three times consecutively meet this characteristic, it is initially determined to be the initial adhesion of mucus. At this point, it can only indicate that the probe has started to have mucus adhesion, but it cannot be determined whether the mucus has become thick enough to affect data acquisition. Further verification is needed by combining the contact duration. Then, the continuous contact duration of the probe is detected simultaneously. The continuous contact duration of the probe refers to the period from when the probe first detects the contact pressure (greater than 0.1N, determined as the start of contact) to when the pressure drops to 0.1N. The following time interval (determined as the end of contact) is used. The typical contact cycle under drying conditions is a baseline value determined by historical data statistics. Specifically, the contact time between the probe and the chili segment is collected under the past 100 drying conditions (no mucus, normal moisture content of the chili segment), and the average value is taken after removing the maximum and minimum values. The mucus interference elimination unit 3 calculates the current continuous contact time of the probe in real time through the timestamp of the contact status signal. If the time does not exceed 0.3 seconds, it means that even if there is a small amount of mucus, it does not affect the contact efficiency and is not judged as mucus thickening. If the time exceeds 0.3 seconds, the mucus thickening judgment is triggered. This is because the more mucus adheres, the stronger its viscosity, which will make the contact and adhesion time between the probe and the chili segment longer, resulting in the contact... If the contact duration significantly exceeds the typical cycle, it can be determined that the slime has thickened to the point where it may interfere with moisture collection. The slime film will hinder the real-time transmission of moisture signals. Finally, the slime adhesion event marker signal is generated only if the exponential decay characteristic and the contact duration exceeding the typical cycle are met. This dual judgment logic can effectively eliminate false judgment scenarios. Only when both characteristics are met simultaneously will the slime interference elimination unit 3 generate the slime adhesion event marker signal and send the signal synchronously to the subsequent cleaning compensation module and sensor monitoring unit 1. This provides a trigger basis for the subsequent reverse superposition of hysteresis deviation and a judgment condition for activating the self-cleaning command, ensuring that every slime adhesion can be accurately captured and avoiding moisture data distortion caused by slime interference.

[0051] The operation of reverse superposition of hysteresis bias in mucus interference elimination unit 3 specifically includes:

[0052] A mapping model between mucus film thickness and response delay is established. The current mucus film thickness is calculated based on the contact pressure attenuation slope. Then, the hysteresis deviation is calculated based on the mapping model. The time axis is shifted to compensate for the hysteresis deviation. The shift amount is equal to the hysteresis deviation. Simultaneously, a self-cleaning command is activated. High-frequency mechanical vibration is generated by piezoelectric ceramics to peel off the mucus layer on the probe surface. After cleaning is completed, the hysteresis deviation counter is reset.

[0053] Further explanation is needed regarding the specific implementation of the reverse superposition of hysteresis bias in the slime interference elimination unit 3. After the slime adhesion event marker signal is generated by the slime interference elimination unit 3, the operation of reverse superposition of hysteresis bias must be initiated immediately. The slime exuded from the chili pepper segment will form a transparent slime film at the probe detection end. This film will hinder the real-time transmission of the moisture signal, causing the raw moisture data collected by the probe to lag behind the actual moisture change of the chili pepper segment by a certain period of time, i.e., the hysteresis bias. If the raw data is used directly, the actual grid moisture value calculated by the dynamic weight allocation unit 2 will deviate from the reality. Therefore, the interference needs to be eliminated through a process, and the specific implementation is as follows:

[0054] A mapping model between mucus film thickness and response delay was established. This model serves as the core bridge connecting the physical morphology of mucus and signal delay. Its construction relies on extensive experimental data to ensure accuracy. The specific process is as follows: Natural mucus was extracted from chili pepper segments, with components consistent with the mucus released during production. Mucus films with thicknesses of 0.05 mm, 0.1 mm, 0.15 mm, 0.2 mm, 0.25 mm, and 0.3 mm were coated onto standard metal test pieces using a precision coating device, covering the possible mucus thickness ranges in actual production. Three parallel samples were prepared for each thickness to reduce error. Test pieces coated with different mucus film thicknesses were brought into contact with a moisture probe, simulating chili pepper segments. The time difference between the moisture signal collected by the probe and the actual moisture value of the test piece was recorded, i.e., the response delay. The raw data was then input into a data processing module, and a nonlinear fitting method was used to establish a model relating mucus film thickness and response delay. The correlation is positive but not perfectly linear, requiring a quadratic function for fitting. For example, the fitted model can be described as the response delay increasing with the thickness of the mucus film, with an average increase of 0.1 seconds for every 0.05mm increase in thickness. Simultaneously, the model error is calculated; if the error exceeds the limit, intermediate thickness samples are added for refitting, ultimately forming a model that accurately outputs the mapping relationship between known thicknesses and corresponding delays. This model is stored in the model database of the mucus interference elimination unit 3 for later use. The current mucus film thickness is calculated based on the contact pressure attenuation slope, a crucial step in converting the real-time contact signal into physical thickness. It is clear that the contact pressure change rate during mucus adhesion exhibits a quasi-exponential decay characteristic, first rising sharply and then gradually decreasing. The contact pressure attenuation slope during the gradual decrease phase is negatively correlated with the mucus film thickness; the thicker the mucus film, the stronger its viscosity and elasticity, resulting in a slower pressure decrease and a smaller absolute value of the attenuation slope. In specific calculations…

[0055] Pressure data during the slow descent phase, i.e., the process from the pressure peak to the stable value, is extracted from the probe contact status signal, typically lasting 0.2-0.5 seconds. Pressure values ​​are taken every 10 milliseconds, and the ratio of the difference between two adjacent pressure values ​​to the time interval is calculated. The average of these ratios is taken as the contact pressure attenuation slope. The calculated attenuation slope is input into the back-derivation module of the established mapping relationship model. The module outputs the current mucus film thickness value according to the preset correlation rules between the attenuation slope and thickness. After obtaining the current mucus film thickness, the hysteresis deviation is calculated based on the mapping relationship model. This deviation is the core parameter that needs to be back-stacked. The specific process is as follows:

[0056] The calculated mucus film thickness is input into the forward query module of the mapping relationship model. The model will output the corresponding response delay based on the fitted relationship. This response delay is the hysteresis bias. For example, a thickness of 0.2 mm corresponds to a response delay of 0.4 seconds in the model. Therefore, the current hysteresis bias is 0.4 seconds, which means that the raw moisture data collected by the probe is actually the moisture value of the chili pepper segment 0.4 seconds ago. This time difference needs to be eliminated through compensation. The raw moisture data stream is time-axis shifted and compensated according to the hysteresis bias to align the lagging raw data with the actual time node. The raw moisture data stream is time-series data generated by sensor monitoring unit 1 every 0.1 seconds. Each data point has a collection timestamp. The specific compensation process is as follows:

[0057] Starting from the time point when the mucus adhesion event marker signal is generated, all uncompensated raw moisture data is extracted backward. These data all have a 0.4-second lag. The extracted data segments are shifted forward by the lag deviation (0.4 seconds) to ensure that the timestamp of each data point precisely corresponds to the time point of the actual moisture change in the chili pepper segment. After shifting, there will be data gaps at the current timestamp. At this time, a forward filling method is used to fill in the gaps to ensure that the moisture data stream is continuous and uninterrupted. Through this step, the operation of backward superimposing the lag deviation is completed, and the real-time moisture value after mucus interference correction is output. While performing time axis shift compensation, the mucus interference elimination unit 3 will simultaneously activate a self-cleaning command to the sensor monitoring unit 1. This command aims to remove the mucus film on the probe surface from a physical level to avoid the continuous accumulation of mucus leading to an increase in deviation. The specific cleaning process is as follows:

[0058] The self-cleaning command triggers the piezoelectric ceramic plate built into the probe detection end, causing it to generate a high-frequency mechanical vibration of 20kHz. This vibration frequency causes the mucus film and the metal end to resonate, breaking the adhesion between them. At the same time, the small centrifugal force generated by the vibration will throw the peeled mucus debris off the probe surface. The vibration duration is set to 2 seconds. During the vibration, the probe will briefly detach from the chili pepper segment (the contact pressure drops to 0N) to prevent mucus debris from adhering to the surface of the chili pepper segment. After cleaning is completed, the sensor monitoring unit 1 will send a cleaning completion signal to the mucus interference elimination unit 3. The unit will then reset the hysteresis deviation counter, clear the counter value to zero, remove the current hysteresis deviation compensation state, and restore the normal contact acquisition of the probe. This ensures that the cycle can start again when the next mucus adhesion event occurs, always maintaining the accuracy of the moisture data and providing reliable basic data for the subsequent displacement compensation algorithm unit 4 and control linkage unit 5.

[0059] The displacement compensation algorithm unit 4 establishes a probability cloud map of chili pepper segment displacement, specifically including:

[0060] The system collects the location coordinates of chili pepper segments over a continuous time series, trains a displacement trajectory prediction model using a convolutional neural network, and outputs the probability density distribution of chili pepper segments appearing in each region of the grid within a future time window. The operation of associating probe spatial confidence includes linking the real-time coordinates of the probe with the probability density distribution of the displacement probability cloud map. Regions with probability density higher than a set threshold are marked as high confidence regions, while those with lower probability density are marked as low confidence regions. Probes located in high confidence regions are assigned high spatial confidence.

[0061] Further explanation is needed regarding the specific implementation of the displacement compensation algorithm unit in establishing the chili segment displacement probability cloud map. After the mucus interference elimination unit 3 completes hysteresis deviation compensation and probe self-cleaning, and outputs the real-time moisture value after mucus interference correction, the dynamic displacement problem of the chili segments in the dehydration chamber still needs to be addressed. The dehydration chamber dehydrates the chili segments through hot air, and the hot air flow will cause the chili segments to undergo slight displacement within the grid. If the probe still collects data at its initial position, the displacement of the chili segments may cause the probe to fall into an area without chili segments, resulting in invalid data collection. Therefore, the displacement compensation algorithm unit 4 needs to first establish a chili segment displacement probability cloud map to clarify the spatial distribution probability of chili segments in future time periods, providing a basis for subsequent probe data screening and supplementary collection. The specific implementation method is as follows:

[0062] Collecting continuous temporal coordinates of chili pepper segments is the fundamental data source for constructing the displacement probability cloud map. It is crucial to ensure the continuity and spatial accuracy of the data. Data acquisition relies on the depth camera and grid positioning sensor in sensor monitoring unit 1. The depth camera collects 3D point cloud data of chili pepper segments within each grid of the dehydration chamber every 0.5 seconds. X and Y correspond to the grid planar coordinates, and Z corresponds to the vertical height. The centroid coordinates of each chili pepper segment are output synchronously. The grid positioning sensor records the grid number of the chili pepper segment at equal time intervals to avoid coordinate confusion caused by cross-grid displacement. During the acquisition process, the continuous temporal data needs to be preprocessed to remove abnormal displacement points caused by hot air turbulence. Simultaneously, the data is stored in a preset format to form a continuous 30-minute temporal coordinate dataset, covering the displacement patterns of chili pepper segments under different hot air intensities. This provides sufficient samples for subsequent model training. A displacement trajectory prediction model is trained using a convolutional neural network. The core function of this model is to predict the displacement trajectory of chili pepper segments within a future period based on historical temporal coordinates, providing a predictive basis for the probability density distribution. The specific training process is as follows:

[0063] The preprocessed temporal coordinate dataset is used to construct samples based on input features and output labels. The input features are the 3D coordinates of a chili pepper segment at 10 consecutive time points, and the output labels are the 3D coordinates of the same segment at the next 5 time points. All samples are divided into a training set (for model learning) and a validation set (for evaluating model performance) in an 8:2 ratio to ensure consistent distribution of displacement amplitude in the two sets. A CNN architecture of 3 convolutional layers, 2 pooling layers, and 2 fully connected layers is then employed. The convolutional layers extract displacement features from the temporal coordinates using 3×3 convolutional kernels. Each convolutional layer is followed by a max-pooling layer to reduce feature dimensionality and retain key displacement information. The fully connected layers map the extracted features... The model is designed to predict the coordinates of chili pepper segments over the next five time points. The output layer uses a linear activation function. Since the predicted coordinates are continuous, no nonlinear mapping is needed. The optimization objective is the error between the predicted and actual coordinates. Gradient descent is used for iterative training. Every 10 training iterations, the error is evaluated using a validation set. If the error on the validation set does not decrease for three consecutive iterations and is below 5%, training is stopped. If the error is too high, the number of convolutional kernels or the pooling window size needs to be adjusted before retraining. Finally, a predictive model that can stably output the displacement trajectory of chili pepper segments within the future time window is obtained. After model training, the probability density distribution of chili pepper segments appearing in each region of the grid within the future time window is output. This is the key step in transforming discrete predicted trajectories into continuous probability distributions. The specific generation process is as follows:

[0064] The dehydration chamber is divided into 400 small regions (5×5mm) each, each with a unique coordinate range to ensure accurate location of potential chili pepper segments. The coordinates of chili pepper segments from the previous 10 time points are input into the trained prediction model to obtain predicted coordinates for the next 5 time points. This process is repeated 20 times, with the timestamp of the input historical coordinates extended by 0.5 seconds each time, covering predictions under different initial states, resulting in 100 predicted coordinates (20 times × 5 time points), forming a set of predicted trajectories. The number of predicted coordinates contained in each sub-region is counted. The probability density of all 400 sub-regions is calculated using this method, and the results are presented as a heatmap, forming a chili pepper segment displacement probability cloud map. This visually reflects the distribution probability of chili pepper segments within the grid within the future time window. Finally, the probe spatial confidence level is correlated. The core operation is to determine the reliability of the probe-collected data by matching the probe coordinates with the displacement probability cloud map. The specific matching process is as follows:

[0065] The real-time three-dimensional coordinates of each moisture probe are retrieved from sensor monitoring unit 1. These coordinates are based on the same spatial coordinate system as the chili pepper segment location coordinates to ensure matching accuracy. A probability density threshold is set, which is determined through historical data statistics. The minimum probability density value of the past 100 probes collecting valid data is taken as the set threshold. Areas with a probability density higher than the threshold are considered high-confidence areas, where the probability of the probe collecting valid data exceeds 90%. Areas with a probability density lower than the threshold are considered low-confidence areas, where the probability of valid data collection is less than 50%. The real-time probe coordinates are then substituted into the displacement probability cloud map to find the grid sub-region to which the coordinates belong. The corresponding probability density is taken. If the coordinates of probe 01 fall in a sub-region with a probability density of 0.8, which is higher than the threshold, the probe is given a high spatial confidence score and marked as confidence score A. The moisture data collected by the probe is determined to be valid. If the coordinates of a probe fall in a sub-region with a probability density lower than the threshold, it is given a low spatial confidence score and marked as confidence score C. Its data will be automatically removed later. Through this matching process, the displacement compensation algorithm unit 4 can accurately identify reliable probe data, laying the foundation for the subsequent removal of low-confidence data and activation of neighboring probes for supplementary collection. This ensures that even if the chili pepper segment is displaced, the spatial coverage integrity of the moisture data within the grid can still be maintained.

[0066] The control linkage unit 5 handles the clustering of large-sized chili pepper segments, specifically including:

[0067] When the actual moisture value of the grid shows that the moisture content of a cluster of large-sized chili pepper segments exceeds the preset moisture threshold, the target area is located based on the centroid coordinates of the cluster in the vector partition map. The air supply temperature and wind speed of the hot air nozzles are increased. The processing of clusters of small-sized chili pepper segments includes reducing the hot air intensity of the corresponding area based on the centroid coordinates of the cluster when the moisture value reaches the critical threshold, and at the same time activating the rotating tray adjustment mechanism to change the spatial distribution of the chili pepper segments.

[0068] Further explanation is needed regarding the specific implementation method of the control linkage unit for processing large and small chili pepper segment clusters. After the displacement compensation algorithm unit 4 establishes the chili pepper segment displacement probability cloud map, ensures the spatial coverage integrity of the probe data, and the dynamic weight allocation unit 2 has output accurate grid moisture values, the control linkage unit 5 needs to implement differentiated processing for large and small chili pepper segment clusters based on the two types of data. Because large chili pepper segments have a large volume and a slow moisture evaporation rate, if the same hot air parameters are used as for small chili pepper segment clusters, it is easy to cause problems such as excessive moisture in large chili pepper segment clusters and excessive drying in small chili pepper segment clusters. Therefore, it is necessary to adjust the hot air system and mechanical structure in a targeted manner based on the cluster size characteristics and spatial location. The specific implementation method is as follows:

[0069] For processing large clusters of chili pepper segments, the core logic is to enhance the effect of hot air to accelerate evaporation when moisture content is high. The first step is to determine whether the moisture content exceeds the standard. The control linkage unit 5 retrieves the actual moisture value of the grid output by the dynamic weight allocation unit 2 in real time, and simultaneously reads the preset moisture threshold set for large clusters of chili pepper segments. This threshold is determined based on the chili pepper processing requirements and the evaporation characteristics of large clusters of chili pepper segments, and is usually 3%-5% higher than the threshold for small clusters of chili pepper segments. That is, when the moisture value of large clusters of chili pepper segments exceeds 20%, it is determined that a state requiring enhanced processing is needed. The three-dimensional coordinates of the centroid of the large cluster of chili pepper segments are extracted from the vector partition map. Because the hot air nozzles in the dehydration chamber are bound according to a grid pattern, each nozzle is responsible for a fixed area of ​​50×50mm. The nozzle number corresponds one-to-one with the grid coordinates. For example, "Nozzle A2" corresponds to the area of ​​X:20-70mm and Y:40-90mm. The system will match the corresponding hot air nozzle according to the centroid coordinates to achieve precise positioning of the target area and avoid affecting the cluster of other areas during adjustment. The control linkage unit 5 sends adjustment commands to the matched hot air nozzle to increase the air supply temperature and wind speed of the hot air nozzle. For example, the original temperature is increased from 50℃ to 55℃, with the temperature increase not exceeding 8℃ to avoid scorching the chili pepper pieces. The original wind speed is increased from 2m / s to 2m / s. The wind speed is controlled at 0.5 m / s, with the increase controlled between 0.3-0.8 m / s to prevent excessive wind speed from causing uncontrolled displacement of the chili pepper segments. Simultaneously, the temperature and wind speed sensors built into the nozzles provide real-time feedback on the adjusted parameters. If the actual temperature does not reach 55℃, the heating power of the nozzles is further fine-tuned to ensure accurate hot air parameters. This intensifies the hot air action, accelerating the evaporation of moisture from large clusters of chili pepper segments until their moisture content drops below a preset threshold. For small clusters of chili pepper segments, mechanical adjustments optimize their spatial distribution to ensure uniform heating and determine if the moisture content has reached the critical threshold. Small clusters of chili pepper segments, due to their small size and rapid evaporation rate, are particularly vulnerable. Excessive drying can cause chili segments to become brittle and lose flavor. Therefore, the critical threshold is set to be 15% lower than that of the large-sized chili segment cluster. When the control linkage unit 5 detects that the moisture value of the small-sized chili segment cluster drops to 15%, it triggers the weakening process to reduce the hot air intensity in the corresponding area. Similarly, based on the centroid coordinates of the small-sized chili segment cluster in the vector partition map, it matches the hot air nozzle responsible for that area and sends a parameter reduction command to it, reducing the air supply temperature from 50℃ to 45℃ and the wind speed from 2m / s to 1.5m / s. If the moisture content continues to drop to 13%, which is below the critical threshold, the temperature is further reduced to 42℃ and the wind speed is further reduced to 1.At a speed of 2 m / s, the system slows down the evaporation rate by reducing the intensity of the hot air, preventing small chili pepper clusters from becoming overly dry. Simultaneously, to prevent uneven hot air distribution from causing some small chili pepper clusters to become too dry, the system compares the moisture content of different small chili pepper clusters within the same nozzle's area of ​​responsibility in real time. If the difference exceeds 2%, the system fine-tunes the nozzle's airflow angle and activates the rotating tray adjustment mechanism to change the spatial distribution of the chili pepper clusters. The rotating tray adjustment mechanism is a rotatable tray built into the bottom of the dehydration chamber grid. Its function is to break up the accumulation of small chili pepper clusters by rotating. Small chili pepper clusters are easily piled up by the hot air, and the accumulated areas evaporate moisture slowly, easily leading to localized over-drying or over-wetting. The control linkage unit 5 determines the tray's rotation parameters based on the centroid coordinates of the small chili pepper clusters. For example, when the centroid is biased to the left of the grid, the tray rotates 15° clockwise; when the centroid is biased to the right, it rotates 15° counterclockwise. The rotation frequency is set to 2 m / s. If the rotation frequency is too high, the chili segments may shift excessively; if it is too low, the clusters cannot be effectively broken up. During rotation, the spatial distribution of the chili segments is monitored in real time using a displacement probability cloud map. If the clustering improves, the rotation frequency is reduced to once per minute. Through the synergistic effect of mechanical adjustment and reduced hot air, the small-sized chili segment clusters are ensured to maintain a uniform spatial distribution and good quality while meeting moisture requirements. Throughout the process, the control linkage unit 5 continuously collects the actual moisture value of the grid and the equipment operating parameters, forming a closed-loop control. If the moisture content of the large-sized chili segment cluster drops to 18%, below the preset threshold of 20%, the hot air parameters are adjusted back to the initial state. If the moisture content of the small-sized chili segment cluster rises to 16%, above the critical threshold of 15%, the rotating tray is paused and the hot air intensity is appropriately increased. This ensures that the moisture content and spatial distribution of both types of clusters remain within the process requirements, guaranteeing the consistency and quality stability of the chili segment processing.

[0070] The logic for generating the mold growth risk signal in the control linkage unit 5 specifically includes:

[0071] The slope of the hysteresis deviation output by the slime interference elimination unit 3 is calculated in real time. When the slope continues to increase, it indicates that the slime adhesion is accelerated. The spatial confidence decay rate of the displacement compensation algorithm unit 4 is analyzed simultaneously. When the rate exceeds the stable working condition threshold, it represents the risk of displacement runaway. The two are normalized and superimposed into a comprehensive risk index. When the index exceeds the dynamic tolerance threshold, the spatial coordinates of the mold risk are output in combination with the vector partition map, and the global hot air equalization system of the dehydration chamber is triggered.

[0072] Further explanation is needed regarding the specific implementation method of generating the mold growth risk signal by the control linkage unit 5. After the control linkage unit 5 completes the precise control and spatial distribution optimization of the hot air parameters for large and small chili pepper clusters, it also needs to monitor and predict the risk of mold growth in real time. In the chili pepper processing scenario, mold growth depends on a high-humidity, poorly ventilated environment. Accelerated adhesion of mucus will cause a local high-humidity layer to form in the contact area between the probe and the chili pepper. The mucus contains polysaccharides, which easily adsorb and lock in moisture, causing the local humidity to rise above 75%. Uncontrolled displacement of the chili peppers will cause cluster accumulation. Hot air cannot penetrate the accumulation area, and the temperature is maintained in the suitable growth range of 25-30℃ for mold. The combination of these two factors will significantly shorten the mold growth cycle. The specific implementation method for generating the risk signal is as follows:

[0073] The slope of the hysteresis deviation output by the slime interference elimination unit 3 is calculated in real time. This slope is a core indicator reflecting the slime adhesion speed. The hysteresis deviation increases with the increase of slime film thickness. If the slope continues to increase, it indicates that the slime adhesion speed is accelerating and the risk of local high humidity is increasing. The specific calculation process relies on the real-time data of the slime interference elimination unit 3, which outputs the hysteresis deviation every 10 seconds. During the calculation, the hysteresis deviation data of 5 consecutive time points are selected, with a time span of 40 seconds, to ensure that the trend change can be captured and to avoid interference from single-point data fluctuations. For example, t1 (10:00:00, 0.4 seconds), t2 (10:00:10, 0.5 seconds), t3 (10:00:20, 0.65 seconds), t4 (10:00:30, 0.85 seconds), and t5 (10:00:40, 1.1 seconds) are selected, and the difference in deviation between two adjacent time points is calculated. (t2-t1=0.1 seconds, t3-t2=0.15 seconds, t4-t3=0.2 seconds, t5-t4=0.25 seconds), each difference is divided by the corresponding time interval (10 seconds) to obtain 4 single slope values ​​(0.01 seconds / second, 0.015 seconds / second, 0.02 seconds / second, 0.025 seconds / second). Finally, the average of these 4 single slope values ​​is taken as the current slope of the hysteresis deviation change. If the slope values ​​calculated for 3 consecutive times show an increasing trend, it is determined that the slime adhesion is accelerated, and the mold risk in the corresponding area needs to be focused on. Simultaneously analyze the spatial confidence decay rate of displacement compensation algorithm unit 4. This rate can intuitively reflect whether the displacement of the pepper segment is out of control. The higher the spatial confidence, the better the position matching between the probe and the pepper segment, and the more stable the displacement. If the decay rate exceeds the threshold, it indicates that the displacement speed of the pepper segment is accelerated and it is easy to accumulate. The specific analysis process is as follows:

[0074] First, spatial confidence data of all probes within a certain grid area is retrieved from displacement compensation algorithm unit 4. The average confidence of the probes within the area is taken as the overall spatial confidence of that area. For example, at time t0 (10:00:00), the overall spatial confidence is 0.9 (high confidence state), and at time t60 (10:01:00), the overall spatial confidence drops to 0.6. Then, the decay rate is calculated, that is, the initial confidence (0.9 at time t0) is subtracted from the current confidence (0.6 at time t60) to obtain the confidence decay amount (0.3), which is then divided by the time interval (60 seconds). That is, the decay rate = 0.3 ÷ 60 = 0.005 / second. The stable operating condition threshold here is a benchmark value determined by statistical analysis of historical data. The spatial confidence decay rate during normal production (no displacement runaway, no accumulation) is used as the maximum value as a threshold (e.g., 0.003 / second). If the current decay rate (0.005 / second) exceeds this threshold, it indicates a risk of displacement runaway, and attention should be paid to the potential for mold growth caused by accumulation. After obtaining the slope of the hysteresis deviation change and the spatial confidence decay rate, both need to be normalized before being combined into a comprehensive risk index. Since the two parameters have different dimensions and numerical ranges, direct superposition will mask the influence of one parameter. Therefore, normalization is the key to ensuring a balanced weighting of the two. During normalization, the historical extreme value ranges of the two parameters are first determined. Based on the statistical data of the past 30 days of production, the slope of the hysteresis deviation change... The historical maximum value of the rate is 0.1 seconds / second, and the minimum value is 0; the historical maximum value of the spatial confidence decay rate is 0.01 seconds / second, and the minimum value is 0. The current parameter values ​​are then mapped to a normalized interval of 0-1. The mapping formula is (current value - minimum value) divided by (maximum value - minimum value). For example, if the current lag deviation change slope is 0.05 seconds / second, the normalized value = (0.05 - 0) ÷ (0.1 - 0) = 0.5; if the current spatial confidence decay rate is 0.005 seconds / second, the normalized value = (0.005 - 0) ÷ (0.01 - 0) = 0.5. The comprehensive risk index is calculated using equal weighting, i.e., comprehensive risk index = normalized lag deviation change slope + normalized spatial confidence decay rate. The reliability decay rate, as in the example above, is 0.5 + 0.5 = 1.0. Then, it is determined whether the comprehensive risk index exceeds the dynamic tolerance threshold. The dynamic tolerance threshold is not a fixed value, but is dynamically adjusted according to the real-time environment of the dehydration chamber. When the overall temperature of the dehydration chamber is below 25℃ (slow mold growth) and the overall humidity is below 65%, the threshold is set to 1.2 (relaxed judgment criteria). When the overall temperature is between 25-30℃ and the overall humidity is between 65%-75%, the threshold is set to 0.8 (tightened judgment criteria, as mold easily proliferates at this time). When the overall temperature is above 30℃ and the overall humidity is above 75%, the threshold is set to 0.6 (the strictest standard). If the current comprehensive risk index (1.0) exceeds the dynamic tolerance threshold (e.g., 0.0), the judgment is not made.If 8) is detected, a mold growth risk warning will be triggered. Further determination of the spatial location of the risk and corresponding countermeasures are required. Obtaining the spatial coordinates of the mold risk requires combining a vector partition map. The vector partition map has marked the three-dimensional coordinates of the centroids of all large and small chili pepper clusters. The control linkage unit 5 will calculate the local risk value corresponding to each cluster, which is the sum of the slope of the normalized hysteresis deviation change and the spatial confidence decay rate of the area where the cluster is located. The cluster with the highest local risk value will be selected, and the centroid coordinates of this cluster will be the spatial coordinates of the mold risk. Simultaneously, this coordinate will be marked with a flashing red indicator on the vector partition map to visually indicate the risk location. Finally, the system will trigger the global hot air equalization system of the dehydration chamber. This system is a global system for the dehydration chamber. The environmental control module eliminates localized high humidity and buildup by synchronously adjusting the airflow parameters and direction of all hot air nozzles. Firstly, it uniformly raises the airflow temperature of all nozzles to 52℃, exceeding the upper limit of the suitable temperature for mold growth, thus inhibiting mold spore germination. The airflow velocity is adjusted to 2.2m / s to enhance airflow penetration and break up clusters of accumulated chili pepper segments. Secondly, it controls the airflow angle of the nozzles, adjusting it from vertically downwards to a 30° tilt to ensure stronger airflow circulation in high-risk areas. Simultaneously, it monitors humidity changes in high-risk areas in real time using humidity sensors. Once the humidity drops below 65%, the hot air parameters are gradually adjusted back to normal operating conditions. Through balanced global control, it effectively blocks mold growth from an environmental perspective, ensuring the processing quality of the chili pepper segments.

[0075] In this invention, sensor monitoring unit 1 collects the moisture value, position coordinates, and probe contact signal of chili segments in the dehydration chamber grid in real time; dynamic weight allocation unit 2 identifies clusters of chili segments of different sizes and assigns corresponding weights to generate the true moisture value of the grid; mucus interference elimination unit 3 marks the adhesion of mucus and compensates for hysteresis deviation; displacement compensation algorithm unit 4 establishes a displacement probability cloud map to realize probe repositioning; and control linkage unit 5 regulates hot air parameters and outputs mold risk signals, thus solving the problems of inaccurate moisture monitoring and difficulty in eliminating interference, and improving processing quality.

[0076] The second objective of this invention is to refer to Figure 2 As shown, a cloud platform is provided for implementing an IoT monitoring system for a chili processing production line, including any of the above-mentioned components. The platform includes: an edge access module, a parallel computing module, a strategy generation module, and a device reverse control module, wherein:

[0077] The edge access module collects and compresses data from the dehydration chamber in real time and uploads it. The parallel computing module runs sensor monitoring unit 1, dynamic weight allocation unit 2, mucus interference elimination unit 3, displacement compensation algorithm unit 4, and control linkage unit 5. The strategy generation module integrates the calculation results to generate control instructions and risk warnings. The equipment reverse control module sends instructions to the dehydration chamber and monitors the execution status.

[0078] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An Internet of Things (IoT) monitoring system for a chili segment processing production line, characterized in that: include: The sensor monitoring unit (1) is used to collect the moisture value, spatial coordinates and probe contact status signals of the chili segments in each grid of the dehydration chamber in real time; The dynamic weight allocation unit (2) automatically identifies the spatial partitions of large-size chili segment clusters and small-size chili segment clusters based on the morphological distribution characteristics of chili segments in the grid obtained by machine vision, constructs a dynamic weight model to assign incremental weight coefficients and decremental weight coefficients to the moisture data of large-size chili segment clusters and small-size chili segment clusters respectively, and generates the true value of grid moisture by superimposing spatial partition weights to eliminate the error of uneven evaporation rate caused by morphological differences. The mucus interference elimination unit (3) marks the mucus adhesion event on the probe surface in real time according to the contact pressure change rate and contact time threshold in the probe contact state signal, triggers the cleaning compensation mechanism, superimposes the hysteresis deviation caused by the mucus film in the original moisture data stream, and activates the self-cleaning command to the sensor monitoring unit (1), outputting the real-time moisture value after mucus interference correction. The displacement compensation algorithm unit (4) establishes a probability cloud map of chili segment displacement based on the temporal displacement trajectory of the chili segment position coordinates, synchronously associates the spatial confidence of each probe, and automatically removes the probe data and activates the neighboring probe to fill in the acquisition when the probe falls into the low confidence area of ​​the displacement probability cloud map, so as to maintain the spatial coverage integrity of multi-point data. The control linkage unit (5) processes the large-size chili pepper cluster and the small-size chili pepper cluster according to the actual value of grid moisture and the spatial partition weight coefficient, and analyzes the slope of the change of hysteresis deviation and the decay rate of spatial confidence in real time. When the superposition of the two exceeds the tolerance, it outputs a mold growth risk signal.

2. The Internet of Things monitoring system for chili segment processing production line according to claim 1, characterized in that: In the dynamic weight allocation unit (2), the spatial partitioning of large-sized chili pepper segment clusters and small-sized chili pepper segment clusters is automatically identified. Specifically, the multi-angle surface texture images and three-dimensional point cloud data of chili pepper segments in the grid are collected synchronously through the machine vision module. The chili pepper segment contour is segmented at multiple scales using a composite feature extraction algorithm. Based on the HSV color space, the chili pepper segments and the background area are separated. Then, the adhesion area of ​​adjacent chili pepper segments is topologically segmented through edge curvature analysis. Finally, the minimum circumscribed cube volume of each chili pepper segment is calculated by combining the Z-axis coordinate data obtained by the depth camera. When the volume value is greater than the preset size threshold, it is classified as a large-sized chili pepper segment cluster. When it is less than the threshold, it is classified as a small-sized chili pepper segment cluster. A vector partition map containing the centroid coordinates of the two types of clusters is generated in real time.

3. The Internet of Things monitoring system for chili segment processing production line according to claim 2, characterized in that, The method for constructing the dynamic weight model in the dynamic weight allocation unit (2) specifically includes: Based on the vector partition map, the proportion of large-sized chili pepper clusters to small-sized chili pepper clusters is statistically analyzed. A cluster size distribution ratio function is established, and the weight coefficient of large-sized chili pepper clusters is set to increase positively correlated with the proportion of their quantity, while the weight coefficient of small-sized chili pepper clusters is set to decrease negatively correlated with their quantity. At the same time, a real-time evaporation rate compensation factor is introduced, and the slope of the weight coefficient is dynamically adjusted according to the hot air temperature gradient in the dehydration chamber to ensure that large-sized chili pepper clusters receive additional weight increase under low evaporation rate conditions.

4. The Internet of Things monitoring system for chili segment processing production line according to claim 3, characterized in that: In the dynamic weight allocation unit (2), when generating the true value of grid moisture through spatial partition weight superposition, a partition moisture value fusion algorithm is adopted, which specifically includes: Extract the probe moisture values ​​corresponding to the centroid coordinates of each cluster in the vector partition map. Weight the high weight coefficient of the large-sized chili pepper cluster with its moisture value and weight the low weight coefficient of the small-sized chili pepper cluster with its moisture value. Then perform spatial interpolation on the two types of weighted values. Combine Kalman filtering to eliminate hot wind disturbance noise and finally output smooth and continuous grid moisture values.

5. The Internet of Things monitoring system for chili segment processing production line according to claim 1, characterized in that: The mucus interference cancellation unit (3) marks mucus adhesion events based on probe contact status signals, specifically including: The probe contact pressure change rate is monitored in real time. When the pressure change rate shows an exponential decay characteristic of first rising sharply and then falling slowly, it is determined to be the initial adhesion of mucus. At the same time, the continuous contact time of the probe is detected. If the time exceeds the typical contact cycle under drying conditions, the mucus thickening judgment is triggered. When both characteristics are met at the same time, a mucus adhesion event marker signal is generated.

6. The Internet of Things monitoring system for chili segment processing production line according to claim 5, characterized in that: The operation of the reverse superposition of hysteresis deviation in the mucus interference elimination unit (3) specifically includes: A mapping model between mucus film thickness and response delay is established. The current mucus film thickness is calculated based on the contact pressure attenuation slope. Then, the hysteresis deviation is calculated based on the mapping model. The time axis is shifted to compensate for the hysteresis deviation. The shift amount is equal to the hysteresis deviation. Simultaneously, a self-cleaning command is activated. High-frequency mechanical vibration is generated by piezoelectric ceramics to peel off the mucus layer on the probe surface. After cleaning is completed, the hysteresis deviation counter is reset.

7. The Internet of Things monitoring system for chili segment processing production line according to claim 1, characterized in that: The displacement compensation algorithm unit (4) establishes a probability cloud map of chili pepper segment displacement, specifically including: The system collects the location coordinates of chili pepper segments over a continuous time series, trains a displacement trajectory prediction model using a convolutional neural network, and outputs the probability density distribution of chili pepper segments appearing in various regions of the grid within a future time window. The operation of associating probe spatial confidence includes matching the real-time coordinates of the probe with the displacement probability cloud map. Regions with probability density higher than a set threshold are marked as high confidence regions, while those with lower probability density are marked as low confidence regions. Probes located in high confidence regions are assigned high spatial confidence.

8. The Internet of Things monitoring system for chili segment processing production line according to claim 7, characterized in that: The control linkage unit (5) handles the clusters of large-sized chili pepper segments, specifically including: When the actual moisture value of the grid shows that the moisture content of a cluster of large-sized chili pepper segments exceeds the preset moisture threshold, the target area is located based on the centroid coordinates of the cluster in the vector partition map. The air supply temperature and wind speed of the hot air nozzles are increased. The processing of clusters of small-sized chili pepper segments includes reducing the hot air intensity of the corresponding area based on the centroid coordinates of the cluster when the moisture value reaches the critical threshold, and at the same time activating the rotating tray adjustment mechanism to change the spatial distribution of the chili pepper segments.

9. The Internet of Things monitoring system for chili segment processing production line according to claim 7, characterized in that: The logic for generating the mold growth risk signal in the control linkage unit (5) specifically includes: The slope of the hysteresis deviation output by the mucus interference elimination unit (3) is calculated in real time. When the slope continues to increase, it indicates that the mucus adhesion is accelerated. The spatial confidence decay rate of the displacement compensation algorithm unit (4) is analyzed synchronously. When the rate exceeds the stable working condition threshold, it represents the risk of displacement runaway. The two are normalized and superimposed into a comprehensive risk index. When the index exceeds the dynamic tolerance threshold, the spatial coordinates of the mold risk are output in combination with the vector partition map, and the dehydration chamber global hot air equalization system is triggered.

10. A cloud platform for implementing the Internet of Things (IoT) monitoring system for a chili segment processing production line as described in any one of claims 1-9, characterized in that, include: The module comprises an edge access module, a parallel computing module, a policy generation module, and a device reverse control module, among which: The edge access module collects and compresses data from the dehydration chamber in real time and uploads it. The parallel computing module runs the sensor monitoring unit (1), dynamic weight allocation unit (2), mucus interference elimination unit (3), displacement compensation algorithm unit (4), and control linkage unit (5). The strategy generation module integrates the calculation results to generate control instructions and risk warnings. The equipment reverse control module sends instructions to the dehydration chamber and monitors the execution status.

Images