A mulch paper monitoring data processing system based on deep Q network

Abstract

The application provides a mulch paper monitoring data processing system based on a deep Q network, relates to the technical field of image processing, and is applied to a farmland monitoring area covered with degradable mulch paper. The system divides a mulch paper monitoring unit, collects multi-period images, space, time and field environment data, performs pixel-level feature extraction, generates mulch paper pixel likelihood values, combines interference masks, covering state masks, soil exposure pixel masks and hole candidate pixel masks, acquires color attenuation, texture degradation, edge curling, hole expansion, covering decline and soil exposure coefficients, generates a mulch state change index, combines state change increments, hole expansion and soil exposure changes, generates a mulch degradation time sequence evolution coefficient, reconstructs a deep Q network state vector, and outputs a target processing action and a monitoring task. The application can improve the accuracy of mulch paper degradation identification, residual risk judgment and review task allocation.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a data processing system for monitoring mulch film based on a deep Q-network. Background Technology

[0002] Plastic mulch film is an agricultural covering material used in farmland mulching, typically laid on the surface of crop planting ridges or sowing strips to improve the soil environment during the early stages of crop growth. After mulching, it reduces soil moisture evaporation, maintains ridge surface humidity, mitigates direct erosion of the topsoil by rainfall, and inhibits weed growth to some extent. Simultaneously, it isolates crop seedlings from some external disturbances, maintaining relatively stable soil temperature and moisture conditions, thus promoting crop emergence, establishment, and early growth. For biodegradable plastic mulch film, after fulfilling its function of covering and conserving moisture, it will gradually undergo changes in color, texture, strength, edge curling, hole formation, and reduced coverage area due to changes in field sunlight, temperature, humidity, rainfall, soil microbial activity, and crop growth. In actual farmland management, the coverage status and degradation process of the plastic mulch film require continuous monitoring. If the plastic film mulch breaks prematurely in the early stages of crop growth, it will increase the exposed soil area, reducing its moisture retention and weed control effects. If the plastic film mulch remains largely intact during its degradable phase, it may affect subsequent tillage, recycling, or incorporation. If the broken plastic film mulch forms localized fragments, it may also increase the risk of residue. Therefore, monitoring the integrity of the plastic film mulch, the extent of damage, the degradation process, and the risk of residue is a crucial aspect of farmland mulch material management.

[0003] Current methods for monitoring plastic film mulch rely heavily on manual inspections, fixed-period photography, or single-image recognition. Manual inspections are limited by field size, crop shading, and inspection experience, making it difficult to continuously record changes in the same area over multiple monitoring periods. While fixed-period photography can acquire image data, it typically only provides static identification of damaged, exposed, or covered areas in the current image, lacking joint analysis of color changes, texture changes, edge curling, hole expansion, and soil exposure trends in the same plastic film area at different times. Ordinary image recognition methods are also easily affected by light reflection, soil adhesion, crop shading, image blurring, and differences in field background, resulting in unstable differentiation between normal degradation, abnormal premature breakage, localized delayed degradation, and residual risk.

[0004] Furthermore, the state changes of biodegradable mulch film exhibit distinct temporal characteristics. Degradation results cannot be judged solely based on a single image from a monitoring period; rather, analysis should consider the continuous trends across multiple monitoring periods. For example, the appearance of holes in a particular area does not necessarily indicate an anomaly. If the rate of hole expansion is consistent with the normal degradation process of the material, it can be considered normal degradation. Conversely, if the hole area and exposed soil area increase rapidly in early monitoring periods, it is more likely to be an abnormal premature rupture. If the color, texture, and coverage area show little change in later monitoring periods, it may indicate localized delayed degradation. Summary of the Invention

[0005] The purpose of this invention is to provide a data processing system for monitoring plastic film based on deep Q-networks, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A data processing system for monitoring agricultural mulch film based on deep Q-networks, applied to monitoring farmland covered with biodegradable agricultural mulch film, includes:

[0008] The monitoring data acquisition module is used to divide the plastic film monitoring units and collect plastic film image data, spatial location data, collection time data and field environment data in multiple monitoring periods to form a multi-period plastic film monitoring dataset.

[0009] The degradation feature calculation module is used to calculate the pixel likelihood value of the mulch film based on the multi-period mulch film monitoring dataset, and combine it with the interference mask, the coverage state mask, the soil exposure pixel mask and the hole candidate pixel mask to obtain the color attenuation coefficient, texture degradation coefficient, edge curling coefficient, hole expansion coefficient, coverage reduction coefficient and soil exposure coefficient, and then generate the mulch film status change index.

[0010] The temporal evolution assessment module is used to generate the temporal evolution coefficient of mulch film degradation based on the mulch film state change index, state change increment, pore expansion coefficient and soil exposure coefficient, and to determine the degradation state.

[0011] The state space construction module is used to generate deep Q-network state vectors based on the mulch film state change index, mulch film degradation time-series evolution coefficient, monitoring reliability coefficient, environmental correction coefficient, and residual risk index.

[0012] The deep Q-network decision module is used to output the action value index corresponding to the data processing action based on the state vector of the deep Q-network, and to determine the target processing action.

[0013] The processing result output module is used to output degradation partitioning results and monitoring tasks based on the target processing actions.

[0014] Compared with the prior art, the beneficial effects of the present invention are:

[0015] This invention improves the accuracy of distinguishing effective pixels of the mulch film, exposed bare soil pixels, and candidate holes by jointly processing the likelihood value of the mulch film pixels with the crop shading pixel mask, soil adhesion pixel mask, reflective interference pixel mask, coverage mask, and effective determination area mask. This significantly reduces the probability of misidentification caused by crop shading, soil adhesion, and strong reflection.

[0016] This invention converts the color attenuation coefficient, texture deterioration coefficient, edge curling coefficient, porosity expansion coefficient, coverage reduction coefficient, and soil exposure coefficient into a mulch film state change index. Furthermore, it combines the state change increment, porosity expansion change, and soil exposure change within a continuous monitoring period to generate a mulch film degradation time-series evolution coefficient. This improves the stability of determining normal degradation, abnormal premature rupture, and local delayed degradation, and significantly enhances the ability to continuously identify the degradation process of mulch film.

[0017] This invention constructs a deep Q-network state vector by incorporating the mulch film state change index, mulch film degradation time-series evolution coefficient, monitoring reliability coefficient, environmental correction coefficient, and residual risk index, and outputs target processing actions based on the action value index. This improves the intelligence level of abnormal area early warning, residual risk marking, image enhancement, adjacent period fusion, and verification sampling task allocation, and significantly enhances the reliability of farmland mulch film monitoring results and the efficiency of subsequent management. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall technology of the present invention;

[0019] Figure 2 This is a schematic diagram illustrating the feature calculation process of the present invention;

[0020] Figure 3 This is a schematic diagram of the timing fireworks evaluation process of the present invention;

[0021] Figure 4 This is a schematic diagram comparing the monitoring results of the present invention. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Please see Figures 1 to 4 This invention provides a technical solution: a data processing system for monitoring agricultural mulch film based on deep Q-networks, applied to monitoring farmland covered with biodegradable agricultural mulch film, comprising:

[0025] The monitoring data acquisition module is used to divide the plastic film monitoring units and collect plastic film image data, spatial location data, collection time data and field environment data in multiple monitoring periods to form a multi-period plastic film monitoring dataset.

[0026] The degradation feature calculation module is used to calculate the pixel likelihood value of the mulch film based on the multi-period mulch film monitoring dataset, and combine it with the interference mask, the coverage state mask, the soil exposure pixel mask and the hole candidate pixel mask to obtain the color attenuation coefficient, texture degradation coefficient, edge curling coefficient, hole expansion coefficient, coverage reduction coefficient and soil exposure coefficient, and then generate the mulch film status change index.

[0027] The temporal evolution assessment module is used to generate the temporal evolution coefficient of mulch film degradation based on the mulch film state change index, state change increment, pore expansion coefficient and soil exposure coefficient, and to determine the degradation state.

[0028] The state space construction module is used to generate deep Q-network state vectors based on the mulch film state change index, mulch film degradation time-series evolution coefficient, monitoring reliability coefficient, environmental correction coefficient, and residual risk index.

[0029] The deep Q-network decision module is used to output the action value index corresponding to the data processing action based on the state vector of the deep Q-network, and to determine the target processing action.

[0030] The processing result output module is used to output degradation partitioning results and monitoring tasks based on the target processing actions.

[0031] Figure 1 The exhibition showcases a farmland monitoring area covered by biodegradable mulch film, mulch film monitoring units, multi-period image acquisition equipment, field environment acquisition devices, and data processing terminals. It uses a technology roadmap to demonstrate the core processing flow of the monitoring data acquisition module, degradation feature calculation module, time-series evolution assessment module, and deep Q-network decision-making module, illustrating the overall technical structure from field data collection to intelligent decision output. Figure 1The diagram shows the farmland monitoring area, mulch film ridge surface, monitoring unit division, and image acquisition devices, spatial location acquisition devices, acquisition time information, and field environment acquisition devices. These are used to represent the collection of mulch film image data, spatial location data, acquisition time data, and field environment data over multiple monitoring periods, forming a multi-period mulch film monitoring dataset. The pixel-level analysis interface, mask processing interface, and feature calculation interface within the data processing terminal in the attached diagram correspond to the degradation feature calculation module. This indicates that the system calculates the pixel likelihood value of the mulch film based on the multi-period mulch film monitoring dataset and, combined with interference masks, cover state masks, soil exposure pixel masks, and hole candidate pixel masks, obtains the color attenuation coefficient, texture degradation coefficient, edge curling coefficient, hole expansion coefficient, cover reduction coefficient, and soil exposure coefficient, thereby generating the mulch film state change index. The multi-period sequence analysis and evolution trend processing interface in the attached diagram corresponds to the time-series evolution evaluation module. This indicates that the system generates the mulch film degradation time-series evolution coefficient and determines the degradation state based on the mulch film state change index, state change increment, hole expansion coefficient, and soil exposure coefficient. The deep Q-network decision interface and output result interface in the attached figure correspond to the state space construction module, the deep Q-network decision module, and the processing result output module. They are used to represent how the system generates a deep Q-network state vector based on the mulch film state change index, mulch film degradation time-series evolution coefficient, monitoring reliability coefficient, environmental correction coefficient, and residual risk index, outputs the action value index, determines the target processing action, and finally outputs the degradation zoning results and monitoring tasks.

[0032] To make the technical solution of the present invention clearer and more complete, a data processing system for monitoring mulch film based on a deep Q-network is described below with reference to specific embodiments. The mulch film in this embodiment is biodegradable. In the initial stage of laying, it is mainly used to cover the ridge surface, reduce soil moisture evaporation, mitigate rain erosion, and inhibit weed germination. During crop growth, the mulch film is affected by light, temperature, soil moisture, rainfall, wind, soil adhesion, crop shading, and the material's own degradation properties, gradually exhibiting changes such as color lightening or darkening, loosening of paper fiber texture, edge curling, hole formation, hole expansion, shrinkage of the covered area, and increase in soil exposure area.

[0033] This implementation method does not directly input the monitoring images of plastic film into a conventional classification model. Instead, it first identifies the areas of plastic film, crop shading, soil coverage, reflective interference, holes, and exposed soil at the pixel level. Then, it converts the pixel-level change results over multiple monitoring periods into plastic film state change index, plastic film degradation time-series evolution coefficient, monitoring reliability coefficient, environmental correction coefficient, and residual risk index. Finally, it constructs a deep Q-network state space and outputs the corresponding data processing actions.

[0034] Example 1

[0035] Please see Figures 1 to 4 This embodiment provides a specific implementation of the monitoring data acquisition module and the degradation feature calculation module. In this embodiment, the monitoring data acquisition module includes a partitioning unit, an image data acquisition unit, an environmental data acquisition unit, and a periodic registration unit; the degradation feature calculation module includes a pixel-level feature extraction unit, a mulch film pixel likelihood calculation unit, an interference mask generation unit, a coverage state mask generation unit, a hole connectivity identification unit, and a basic degradation feature calculation unit. The system comprises several modules: a partitioning unit to determine the monitoring units and their unit numbers; an image data acquisition unit to acquire multi-source images; a periodic registration unit to unify images from multiple monitoring periods into the same plot coordinate system; a pixel-level feature extraction unit to extract brightness, chromaticity, near-infrared, edge, and fiber texture features; a mulch film pixel likelihood calculation unit to calculate the pixel likelihood value of a pixel belonging to the mulch film; an interference mask generation unit to generate crop shading pixel masks, soil adhesion pixel masks, and reflective interference pixel masks; a coverage status mask generation unit to generate effective pixel masks for clean mulch film, coverage presence masks, and effective determination area masks; a hole connectivity identification unit to identify candidate hole pixels and generate a set of hole connectivity components; and a basic degradation feature calculation unit to calculate color attenuation coefficient, texture degradation coefficient, edge curling coefficient, hole expansion coefficient, coverage reduction coefficient, and soil exposure coefficient. The system was applied to a 3.6-hectare cornfield covered with biodegradable mulch film, which was divided into several mulch film monitoring units according to the mulch film laying boundary, crop planting ridge direction, and plot spatial boundary. In this embodiment, seven monitoring units of plastic film were selected as sample examples, and were labeled as M01, M02, M03, M04, M05, M06 and M07 respectively.

[0036] The image data acquisition unit uses a UAV multispectral camera to acquire top-view images of the ridge surface, a ground-based fixed camera to acquire low-angle edge images, and a mobile inspection camera to acquire local images of holes. The UAV images are used to determine the area covered by the plastic film and the area of ​​bare soil, the ground-based fixed camera is used to identify edge curling and the width of the curl, and the mobile inspection camera is used to supplement details of hole boundaries, paper fiber breaks, and soil adhesion. The periodic registration unit unifies the images from days 7, 14, 21, and 28 after mulching into the same plot coordinate system, and completes pixel-level registration using the intersection of ridge lines, the corner points of the plot boundaries, and the initial mulch film outline. Considering that the canopy will obscure the visual features of the ground surface during the mid-to-late stages of crop growth (such as day 21 and day 28), this system adopts a two-step registration method: first, large-scale coordinate system registration is completed using the device's built-in RTK (real-time dynamic differential positioning) or a fixed calibration board in the field; then, fine pixel-level alignment calibration is performed in the gaps / visible areas in the image that are not obscured by leaves, using ridge lines or the initial outline.

[0037] The system denotes the pixel set of the i-th monitoring unit as Ωᵢ and the pixel coordinates as p=(x,y). To distinguish between mulch film, crops, soil, reflections, and holes, the system first converts the RGB image to the Lab and HSV color spaces and calculates the vegetation index using the near-infrared channel. Biodegradable mulch film typically has a light-colored surface, fine fibrous lines, continuous edge lines, and brightness and chromaticity distributions different from bare soil; crop pixels have a high vegetation index, soil pixels have a brownish-red low-brightness characteristic, and reflective pixels have a high-brightness, low-saturation characteristic. Specifically, the pixel-level feature extraction unit performs pixel-level feature extraction on the p-th pixel of the i-th mulch film monitoring unit within the t-th monitoring period, and calls the brightness similarity subunit, chromaticity similarity subunit, near-infrared molecular unit, edge continuity subunit, and paper fiber texture subunit respectively to obtain five types of pixel-level recognition scores. Further, the mulch film effective pixel recognition unit performs pixel-level feature extraction on the p-th pixel of the i-th mulch film monitoring unit within the t-th monitoring period. The monitoring unit for plastic film was in the first The first monitoring cycle Pixel-level feature extraction was performed on each pixel, and brightness similarity score, chromaticity similarity score, near-infrared discrimination score, edge continuity score, and paper fiber texture score were calculated respectively. All five scores are normalized values, ranging from 0 to 1. The closer the value is to 1, the more the pixel matches the effective pixel characteristics of biodegradable mulch film; the closer the value is to 0, the more likely the pixel belongs to an area obstructed by soil, crops, mud, shadows, or reflective interference.

[0038] S1011, brightness similarity score is recorded as This value is used to characterize the consistency between the current pixel brightness and the brightness center of the plastic film during the initial laying cycle. In the initial laying cycle, the system selects a reference area of ​​plastic film that is free from crop obstruction, soil cover, and significant reflection. The average L channel value of all plastic film pixels within this reference area in the Lab color space is calculated and used as the initial brightness center of the plastic film, denoted as . ; then read the first The monitoring unit for plastic film was in the first Within the monitoring period, the first The L-channel luminance value of each pixel, denoted as The system normalizes and compares the absolute difference between the two values ​​with a preset brightness tolerance difference to obtain a brightness similarity score. The calculation formula is as follows:

[0039]

[0040] In the formula, Indicates the first The monitoring unit for plastic film was in the first Within the monitoring period, the first The brightness similarity score of each pixel; This represents the L channel luminance value of the current pixel in the Lab color space; Indicates the first The brightness center of the mulch film monitoring unit during the initial laying period; Indicates the allowable difference in brightness; This represents a constant used to prevent the denominator from being zero.

[0041] Preferably, , Because biodegradable plastic film is affected by light, humidity, dust, and material aging during field use, its brightness will naturally drift within a certain range. Therefore, a completely equal brightness matching method is not adopted; instead, an acceptable range is defined by a brightness tolerance. The closer the current pixel brightness is to the initial brightness center of the plastic film, the better. The larger the value; when the brightness difference exceeds the allowable brightness difference. A value close to 0 indicates that the pixel is more likely to be in a shadow, bare soil, highly reflective area, or an area not covered by mulch film.

[0042] S1012, the chromaticity similarity score is recorded as follows: These values ​​are used to characterize the proximity of the current pixel's a and b channels in the Lab color space to the initial chromaticity center of the mulch film. The system first acquires the average a and b channels of the mulch film reference area during the initial laying cycle, denoted as [values ​​to be inserted here]. and Then read the a-channel and b-channel values ​​of the current pixel, and record them as follows: and The system calculates the two-dimensional chromaticity distance between the current pixel and the initial chromaticity center of the mulch film, and normalizes it into a chromaticity similarity score. The calculation formula is as follows:

[0043]

[0044]

[0045] In the formula, Indicates the first The monitoring unit for plastic film was in the first Within the monitoring period, the first The chromatic distance between each pixel and the initial chromaticity center of the mulch film; This represents the a channel value of the current pixel in the Lab color space; This represents the b channel value of the current pixel in the Lab color space; This indicates the a-channel chromaticity center of the base area of ​​the linerboard during the initial laying cycle; This indicates the b-channel chromaticity center of the base area of ​​the liner during the initial laying cycle; Indicates the color similarity score; Indicates the permissible chromaticity distance; This represents a constant to prevent the denominator from being zero. Preferably, , Because biodegradable plastic film may yellow, gray, locally brown, and mottle during the degradation process, brightness alone cannot reliably distinguish between the plastic film, soil, and soil-attached areas. Therefore, the chromaticity distance formed by the a and b channels is used for judgment. When the chromaticity distance of the current pixel is less than the allowable chromaticity distance, it indicates that the pixel is still close to the original color system of the plastic film; when the chromaticity distance increases, it indicates that the pixel may belong to bare soil, crop shadow, soil attachment, or strongly degraded patches.

[0046] S1013, Near-infrared discrimination score is recorded as follows: This is used to distinguish between the pixels of the plastic film and the pixels of crop vegetation using near-infrared reflectance characteristics. The system reads the first... The monitoring unit for plastic film was in the first Within the monitoring period, the first The normalized value of near-infrared reflectance of each pixel, denoted as . Read the near-infrared reflection center of the base area of ​​the membrane paper during the initial laying cycle, and record it as... Simultaneously, based on the crop canopy segmentation results, the near-infrared reflection centers of crop pixels within the current monitoring period are statistically analyzed and recorded as follows: .

[0047] The crop canopy segmentation results are obtained by reading the visible light color information and near-infrared reflectance information of each pixel within the current monitoring period. Since crop leaves typically exhibit a distinct green channel response, and their reflectance values ​​in the near-infrared channel are significantly different from those of the mulch film and bare soil, the system first marks pixels with vegetation spectral characteristics as candidate pixels for the crop canopy. These vegetation spectral characteristics mainly include: strong near-infrared reflectance, a clear green response in visible light, and pixel colors that do not conform to the light-colored surface characteristics of the mulch film or the brownish-red characteristics of bare soil. Then, candidate pixels for the crop canopy are excluded. If a pixel exhibits high brightness and low saturation, it is identified as a reflective interference pixel on the mulch film surface and is not included in the crop canopy count. If a pixel is brownish-red and lacks the near-infrared reflectance characteristics of crop leaves, it is identified as a bare soil pixel or a soil-attached pixel and is not included in the crop canopy count. If a pixel is located within the initial mulch film coverage area and there are still mulch film support features around it, it is preferentially treated as a soil-attached area and is also not included in the crop canopy count. After initial exclusion, the system combines the crop planting row spacing, planting ridge direction, and the current monitoring unit's location to perform spatial consistency judgment on the remaining candidate pixels. In other words, only pixels located near crop planting rows, matching the crop's growth position, and forming continuous leaf areas in the image are retained as crop canopy pixels. Isolated noise points, small false positives, and scattered pixels that do not conform to canopy morphology are removed by the system through connected component filtering and morphological processing, yielding the crop canopy segmentation result for the current monitoring period. The crop canopy segmentation result is essentially the set of all pixels in the current image identified as crop leaf occlusion areas. The system only counts near-infrared reflectance values ​​within this crop canopy pixel set and uses the central value as the near-infrared reflectance center of the crop pixels for the current monitoring period. This near-infrared reflectance center is used to subsequently determine whether a pixel to be identified is closer to a crop leaf or closer to the mulch film, thus avoiding misclassifying crop-occluded areas as valid mulch film pixels.

[0048] Since crop leaves typically have high near-infrared reflectance, and the near-infrared reflectance of mulch film differs from that of crop vegetation, the system simultaneously calculates the proximity of the current pixel to the initial near-infrared center of the mulch film, as well as the separation of the current pixel from the near-infrared center of the crop. The calculation formula is as follows:

[0049]

[0050]

[0051]

[0052] In the formula, This represents the proximity score between the current pixel and the initial near-infrared reflection center of the mulch film; This represents the separation score between the current pixel and the near-infrared reflection center of the crop; Indicates the near-infrared discrimination score; This represents the normalized near-infrared reflectance value of the current pixel; This indicates the near-infrared reflection center of the base area of ​​the membrane paper during the initial laying cycle; Indicates the near-infrared reflection center of the crop pixel within the current monitoring period; Indicates the near-infrared tolerance; Indicates the weight of the near-infrared proximity score; This represents a constant to prevent the denominator from being zero. Preferably, , , Since this parameter is primarily used to identify the effective pixels of the mulch film, whether the current pixel is close to the near-infrared reflectance characteristics of the initial mulch film should be the main criterion. Simultaneously, to exclude areas obscured by crop leaves, a crop near-infrared reflectance separation score is also introduced. When the mulch film is obscured by crop leaves, the near-infrared reflectance is usually closer to the crop canopy than to the mulch film. It will decrease.

[0053] S1014, edge continuity score is denoted as This is used to characterize whether there are continuously extending edge structures along the laying direction of the mulch film within the neighborhood of the current pixel. The system first determines the [missing information - likely a specific value or parameter] based on the boundary line of the mulch film within the initial laying cycle. The directional angle of the mulch film laying in each mulch film monitoring unit is denoted as... Then, using the current pixel... Construct a local pixel window centered on the central pixel, denoted as . The system extracts the set of edge pixels within this local pixel window using the Sobel operator or the Canny edge detection method, denoted as . For each edge pixel Calculate its edge direction angle, denoted as . When edge pixels When the orientation angle satisfies the orientation consistency condition, the edge pixel is counted in the number of edge pixels with consistent orientation. The calculation formula is as follows:

[0054]

[0055]

[0056]

[0057]

[0058] In the formula, Represents edge pixels The determination value for whether it is consistent with the direction of the mulch film laying; Represents edge pixels Edge direction angle; Indicates the first The direction angle of mulch film laying in each mulch film monitoring unit; Indicates the allowable deviation of the orientation angle; Indicates the proportion of edge pixels with consistent direction; Represents the set of edge pixels within a local pixel window; This represents the normalized score of the longest consecutive edge segment; It represents the length of the longest continuous edge segment extending along the laying direction of the mulch film within the local pixel window; Indicates the reference length of continuous edge segments; Indicates continuous edge scores; Indicates the proportional weight of edge pixels with consistent direction; This represents a constant to prevent the denominator from being zero. Preferably, a local pixel window. Set to 15 pixels x 15 pixels, with allowable deviation in orientation angle. Reference length of continuous edge segment 1 pixel, , Because the edges of plastic film, fold boundaries, and crack / hole edges typically have continuous linear features extending along the laying direction or related to the laying boundary, while soil particles, isolated noise, and small-area shadow edges usually lack stable directional continuity, therefore... It can improve the identification stability of the edge area and hole boundary area of ​​the mulch film.

[0059] S1015, the paper fiber texture score is recorded as follows: This is used to characterize whether there are fine paper fiber textures on the surface of degradable mulch film within the neighborhood of the current pixel. The system uses the current pixel... Construct a local texture window centered on the texture window, denoted as . The system extracts grayscale gradient, directional texture, and fine line response within this local texture window. Biodegradable mulch film, before complete rupture, typically exhibits a fine, relatively consistent fiber texture or embossed texture; bare soil usually displays a granular, randomly oriented coarse texture, while crop leaves show distinct vein directions but differ in color and near-infrared characteristics. Therefore, the system uses local directional consistency and fine line response intensity to jointly calculate the paper fiber texture score, specifically including: the system calculates the local texture window... The horizontal and vertical gradients within the gradient are denoted as follows: and The local texture orientation consistency score is calculated based on the structure tensor, as shown in the following formula:

[0060]

[0061] Furthermore, the system employs a multi-directional fine-line filter to extract local texture windows. The response of the fine lines in the paper fibers was analyzed, and the normalized value of the fine line response was obtained, denoted as [value missing]. The paper fiber texture score is calculated using the following formula:

[0062]

[0063] In the formula, Indicates the first The monitoring unit for plastic film was in the first Within the monitoring period, the first The paper fiber texture score per pixel; This indicates the score for local texture orientation consistency. This represents the normalized value of the response to fine lines in paper fibers; Indicates the current pixel A local texture window centered on the subject; Represents any pixel within a local texture window; Represents pixels Horizontal gradient; Represents pixels The vertical gradient; This represents the weight of the local texture orientation consistency score; This represents a constant to prevent the denominator from being zero. Preferably, a local texture window. Set to 17 pixels x 17 pixels. , Since the surface texture of biodegradable mulch film includes both fiber orientation consistency and fine line response intensity, using only orientation consistency may misidentify crop leaf veins as mulch film texture, while using only fine line response intensity may misidentify soil particle texture as paper texture. Therefore, the two are weighted equally to improve the reliability of paper fiber texture recognition.

[0064] S1016. Further, the mulch film pixel likelihood calculation unit receives the brightness similarity score, chromaticity similarity score, near-infrared discrimination score, edge continuity score, and paper fiber texture score, and performs weighted fusion based on the pixel recognition weight configuration result to obtain the pixel likelihood value of pixel p belonging to mulch film in the t-th monitoring period. The pixel recognition weight configuration result includes brightness similarity weight, chromaticity similarity weight, near-infrared discrimination weight, edge continuity weight, and paper fiber texture weight. Based on the above five pixel-level scores, the pixel likelihood value of pixel p belonging to mulch film in the t-th monitoring period is generated, denoted as... The calculation formula is as follows:

[0065]

[0066] In the formula, Indicates the first The monitoring unit for plastic film was in the first Within the monitoring period, the first The comprehensive recognition score of each effective pixel of the mulch film paper; Indicates the brightness similarity score; Indicates the color similarity score; Indicates the near-infrared discrimination score; Indicates continuous edge scores; Indicates the score for paper fiber texture; , , , and This indicates the corresponding weight. Preferably, , , , , Since the chromatic similarity score can distinguish between the mulch film, soil, and mud-attached areas, and the paper fiber texture score can reflect the material surface characteristics that differentiate biodegradable mulch film from bare soil and crop leaves, the chromatic similarity score and paper fiber texture score are given relatively high weights. The near-infrared discrimination score is used to suppress crop shading misidentification and is set to 0.20. The brightness similarity score and edge continuity score are used to supplement the judgment of lighting conditions and laying boundary continuity, respectively, and are set to 0.18. The sum of the above weights is 1.

[0067] S1017. It should be noted that... Only pixels The probabilistic tendency of the mulch film material cannot be directly equated with the final effective pixels of the mulch film. Furthermore, after receiving the likelihood values ​​of the mulch film pixels, the interference mask generation unit does not directly use pixels whose likelihood values ​​reach a threshold as the final effective pixels. Instead, it first combines the vegetation index, HSV high-brightness low-saturation features, soil color likelihood values, and the neighboring mulch film support ratio to generate crop occlusion pixel masks, reflective interference pixel masks, and soil-attached pixel masks, respectively. Based on the interference mask generation results, the cover state mask generation unit generates a clean mulch film effective pixel mask, a cover presence mask, and an effective determination area mask. Thus, crop occlusion pixels and highly reflective pixels are treated as undeterminable pixels, soil-attached pixels are treated as pixels that exist under cover but do not participate in color and texture calculations, and clean mulch film effective pixels are used for color attenuation and texture degradation calculations.

[0068] The system first generates a crop occlusion pixel mask, denoted as . The formula is as follows:

[0069]

[0070] In the formula, This represents a pixel mask that is obscured by crops. Indicates the first The monitoring unit for plastic film was in the first Within the monitoring period, the first The vegetation index corresponding to each pixel; This indicates the crop shading threshold. Preferably, .when When the value is greater than or equal to 0.32, it indicates that the pixel better matches the characteristics of high near-infrared reflectance and green channel enhancement of vegetation, and the system marks it as a crop shading pixel. Crop shading pixels are not included in the calculation of mulch film color decay, texture degradation, holes and soil exposure, and are treated as undeterminable pixels in subsequent cover reduction calculations, rather than being directly treated as mulch film missing pixels.

[0071] The system generates a mask of reflective interference pixels, denoted as . The formula is as follows:

[0072]

[0073] In the formula, Indicates a pixel mask that reflects light and causes interference; Indicates the first The brightness value of each pixel in the HSV color space; Indicates the first The saturation value of each pixel in the HSV color space; Indicates the reflectivity threshold; This represents the reflectivity saturation threshold. Preferably, , When a pixel has high brightness and low saturation, it indicates that the pixel is more likely to be a bright area caused by strong reflection from the surface of the mulch film or camera exposure. Reflective interference pixels are not included in color attenuation and texture degradation calculations and are excluded in hole identification to prevent bright, texture-deficient areas from being misidentified as cracked areas.

[0074] The system generates a pixel mask for soil adhesion, denoted as... Both areas with soil adhesion and areas with exposed bare soil can appear brownish-red. However, areas with soil adhesion are usually located within the initial area covered by the mulch film, appearing as localized patches, and the edges of the mulch film or the supporting information of the paper fibers still exist in their vicinity. Areas with exposed bare soil typically correspond to areas where the mulch film is missing and are adjacent to the boundaries of holes or cracks. Therefore, the system first calculates the pixels. The likelihood value of soil color Then calculate the pixels. The proportion of plastic film support in the neighborhood The formula is as follows:

[0075]

[0076]

[0077] In the formula, Represents pixels The proportion of plastic film support within the adjacent area; Represented in pixels A local window centered on the user; Indicates the number of pixels within a local window; Represents pixels within a local window The pixel likelihood value belonging to the mulch film paper; This represents the pixel likelihood threshold of the plastic film. This indicates the initial mulch film covering the mask; Represents pixels Likelihood values ​​of pixels belonging to the soil-attached region; Indicates the soil adhesion threshold; This indicates the neighboring area's mulch film support threshold. Preferably, , , , According to this formula, only pixels within the initial mulch film coverage area that possess soil color characteristics, are still supported by mulch film, and are not affected by crop shading or reflective interference are marked as soil-attached pixels. Soil-attached pixels are not included in color and texture calculations as valid pixels of clean mulch film, but they are not directly considered as missing mulch film in the mulch presence determination.

[0078] After determining the masking effects of crop shading, light reflection interference, and soil adhesion, the system generates a clean, effective pixel mask for the mulch film, denoted as... The formula is as follows:

[0079]

[0080] In the formula, This indicates the effective pixel mask for the cleaned mulch film. Represents pixels The pixel likelihood value belonging to the mulch film paper; The pixel likelihood threshold for the plastic film is represented, with a preferred value of 0.62. This indicates the initial mulch film covering the mask; This represents a pixel mask that is obscured by crops. This represents a pixel mask indicating soil adhesion. This represents the pixel mask for reflective interference. According to this formula, only pixels within the initial mulch film laying area, whose pixel likelihood value reaches the threshold, and which are not affected by crop shading, soil adhesion, or reflective interference, are counted as effective pixels of the clean mulch film. These effective pixels are used in subsequent calculations of color attenuation coefficients and texture degradation coefficients.

[0081] Furthermore, to avoid mistaking areas with soil adhesion and areas where reflection is indeterminate as ruptured mulch film, the system generates a masking layer for the covering, denoted as... The formula is as follows:

[0082]

[0083] In the formula, This indicates that a mulch film is present. When a pixel is a valid pixel of clean mulch film or a pixel with soil adhesion, the system considers that the pixel location may still have mulch film coverage and does not directly treat it as a hole or exposed bare soil pixel. Pixels with crop shading and strong reflectivity are considered undeterminable pixels and are not included in the denominator statistics for coverage loss to prevent the coverage reduction coefficient caused by crop leaf coverage or reflection from being amplified.

[0084] The system generates a mask for the effective determination region, denoted as . The formula is as follows:

[0085]

[0086] In the formula, This indicates the effective determination area mask. This mask is used to limit the statistical range for subsequent coverage reduction and hole identification; pixels that are blocked by crops or subject to strong reflective interference are temporarily excluded and are not directly included in the area of ​​missing mulch film.

[0087] The system generates a soil exposure pixel mask, denoted as The system first calculates the pixels. Pixel likelihood values ​​belonging to bare soil regions The likelihood value of a pixel is determined by soil color center similarity, low fiber texture response, grain coarse texture response, and spatial adjacency with areas lacking plastic film. When the likelihood of a bare soil pixel reaches a threshold, and the pixel is not affected by crop shading, soil adhesion, or reflective interference, the system marks it as a soil-exposed pixel, as shown in the following formula:

[0088]

[0089] In the formula, Indicates the pixel mask for soil exposure; Represents pixels Likelihood values ​​of pixels belonging to the bare soil region; This represents the threshold value for bare soil pixels. Preferably, This treatment can distinguish between areas of truly bare soil and areas where soil adheres to the surface of the plastic film.

[0090] Furthermore, the hole connectivity identification unit does not directly use black or brown pixel thresholds to identify holes. Instead, it generates a candidate pixel mask for holes under the joint constraints of the initial mulch film cover mask, the effective determination area mask, the cover presence mask, the soil exposure pixel mask, and the likelihood value of the hole pixels. In this way, the hole connectivity identification unit can eliminate the influence of crop shading, soil adhesion, reflective interference, and isolated noise points on hole identification. The system calculates the likelihood value of the hole pixels. The pixel likelihood value is obtained by considering the closure degree of the hole edge, the color consistency of the bare soil, the boundary tear texture, and the continuity of the edges of adjacent plastic film sheets. The candidate pixel mask for the hole is denoted as... The calculation formula is as follows:

[0091]

[0092] In the formula, Represents the candidate pixel mask for holes; This indicates the initial mulch film covering the mask; Indicates the effective determination area mask; This indicates that a mask exists for the coverage; Indicates the pixel mask for soil exposure; This represents the likelihood value of the hole pixel; This represents the pixel threshold for holes. Preferably, This formula identifies only pixels within the initial mulch film coverage area that are not currently covered, can be effectively determined, show signs of soil exposure, and have a hole likelihood value that reaches the threshold as candidate holes. This process avoids misclassifying crop shading, soil adhesion, and reflective interference as holes.

[0093] The system performs opening and closing operations and connective component labeling on the candidate pixel mask for holes, and removes regions with an area smaller than the minimum connected component area of ​​the hole to obtain the set of connected components for the holes, as shown in the following formula:

[0094]

[0095] In the formula, Indicates the first The monitoring unit for plastic film was in the first The set of connected domains of holes within a monitoring cycle; Indicates the first A hole-connected domain; Indicates the first The number of pixels contained in a connected region of a hole; This represents the area of ​​the smallest connected region with a hole. Preferably, Each pixel. This processing can remove mud spots, small shadow patches, and camera noise points, making the hole recognition results more consistent with the actual shape of the mulch film after it is torn.

[0096] S1018. Further, the basic degradation feature calculation unit receives the effective pixel mask of the clean mulch film, the cover presence mask, the effective determination area mask, the soil exposure pixel mask, the set of connected domains of holes, and the current set of edge pixels of the mulch film, and calculates the color attenuation coefficient, texture degradation coefficient, edge curling coefficient, hole expansion coefficient, cover reduction coefficient, and soil exposure coefficient, respectively. Among these, the color attenuation coefficient and texture degradation coefficient are calculated only based on the effective pixels of the clean mulch film; the cover reduction coefficient is calculated based on the cover presence mask and the effective determination area mask; the hole expansion coefficient is calculated based on the number of connected domains of holes, the hole area, and the hole area growth rate; and the soil exposure coefficient is calculated based on the soil exposure pixels within the initial cover area. The number of effective pixels of the clean mulch film in the current monitoring period is calculated and denoted as... The formula is as follows:

[0097]

[0098] The color attenuation coefficient is calculated based on the color difference between the effective pixels of the cleaned mulch film and the initial laid image, using the following formula:

[0099]

[0100] In the formula, Indicates the color attenuation coefficient; This indicates the number of effective pixels in the clean mulch film during the current monitoring period; Represents pixels Color difference in Lab space relative to the initial color of the mulch film; This indicates the preset maximum color difference normalization value; This represents a constant to prevent the denominator from being zero. Preferably, , .

[0101] The texture degradation coefficient is calculated by combining the changes in the gray-level co-occurrence matrix, the changes in the local binary mode, and the crack line density, as shown in the following formula:

[0102]

[0103] In the formula, Indicates the texture degradation coefficient; This represents the change in gray-level co-occurrence matrix contrast and homogeneity relative to the initial mulch film texture. Indicates the difference in histograms of local binary patterns; The crack line density is represented by the ratio of the length of the skeletonized crack pixels to the effective pixel area of ​​the mulch film. , and All values ​​are normalized dimensionless values, ranging from 0 to 1; , and This indicates the texture degradation weight. Preferably, , , .

[0104] The edge curling coefficient is calculated based on the pixel offset distance of the current edge line of the mulch film relative to the initial laying reference edge line, as shown in the following formula:

[0105]

[0106] In the formula, Indicates the edge curl factor; This represents the set of pixels at the edge of the plastic film identified during the current monitoring period. Indicates the initial laying reference edge line; Represents edge pixels The shortest pixel distance to the initial baseline edge; Indicates the first The pixel value corresponding to the initial width of the mulch film of each monitoring unit; This indicates the current number of edge pixels, and ; This represents a constant used to prevent the denominator from being zero.

[0107] The cavity expansion coefficient is calculated based on the number of cavities, the cavity area, and the cavity area growth rate, using the following formula:

[0108]

[0109] In the formula, Indicates the hole expansion coefficient; Indicates the number of connected domains of holes in the current monitoring period; This represents the normalized upper limit of the number of holes; This represents the total area of ​​the current hole pixels; This represents the total area of ​​the holes in the previous monitoring period; Indicates the first Total pixel area of ​​each mulch film monitoring unit; , and Indicates the hole expansion weight; This represents a constant to prevent the denominator from being zero. Preferably, , , , .

[0110] The cover degradation coefficient is not solely based on the effective pixels of the clean mulch film, but rather on the presence of a masking layer. For areas obstructed by crops and areas with strong reflective interference, the system excludes them from the statistical denominator as undeterminable areas; for areas with soil adhesion, the system includes them in the coverage area to avoid misjudging mud on the mulch film surface as a lack of coverage. The formula for calculating the cover degradation coefficient is as follows:

[0111]

[0112] In the formula, Indicates the coverage reduction factor; This indicates that a mask exists for the coverage; Indicates the effective determination area mask; This indicates the initial mulch film covering the mask; This represents a constant to prevent the denominator from being zero. Using this formula, the system calculates cover reduction only within the identifiable area, avoiding incorrect inclusion of crop shading and highly reflective areas in the calculation of mulch film loss.

[0113] The soil exposure coefficient is calculated based on the soil exposure pixels within the initial coverage area, using the following formula:

[0114]

[0115] In the formula, Indicates the soil exposure coefficient; Indicates the pixel mask for soil exposure; This indicates the initial mulch film covering the mask; Indicates the effective determination area mask; This represents a constant to prevent the denominator from being zero. This formula only counts exposed bare soil pixels within the initial mulch film coverage area, directly linking the soil exposure coefficient to mulch film tears or missing coverage. Pixel-level recognition samples from Example 1 are shown in Table 1 below.

[0116] Table 1: Pixel-level recognition examples of seven mulch film monitoring units

[0117] M01 120000 74600 16800 0.08 5 Continuous paper surface with small holes in some areas M02 120000 49200 38600 0.11 14 Numerous holes with obvious outward expansion M03 120000 91400 8200 0.06 2 The plastic film remains intact, and its degradation is relatively slow. M04 120000 35800 51200 0.09 18 Concentration of cracks and exposed bare soil M05 120000 68200 21400 0.17 6 Crop shading affects local assessment M06 120000 70200 18400 0.24 5 The image is heavily obscured and needs to be reviewed. M07 120000 54400 34200 0.10 13 The number of holes and edge curling increases simultaneously.

[0118] As shown in Table 1, the effective pixels of the clean mulch film in M04 are significantly reduced, while the number of exposed soil pixels and the number of connected voids are both high, consistent with the image characteristics of continuous bare soil and broken edges after premature rupture of the biodegradable mulch film. The effective pixels of the clean mulch film in M03 are still relatively high, but the number of connected voids is low, requiring further analysis based on the monitoring period to determine if it represents localized delayed degradation. The crop shading ratio in M06 reaches 0.24, indicating a large number of undetermined pixels in its image; further analysis should consider the monitoring reliability coefficient to determine whether to perform a re-sampling. Through the aforementioned pixel-level identification and mask generation process, the system can separate the mulch film surface, soil adhesion, ruptured voids, exposed bare soil, crop shading, and reflective interference, providing a clear data source for subsequent calculations of color attenuation coefficient, texture degradation coefficient, edge curling coefficient, void expansion coefficient, cover reduction coefficient, and soil exposure coefficient.

[0119] Example 2

[0120] This embodiment, based on Embodiment 1, further illustrates how the degradation feature calculation module converts pixel-level recognition results into a mulch film state change index. The system uses a weighted coupling calculation algorithm to obtain the mulch film state change index, denoted as DMSI, for color attenuation coefficient, texture degradation coefficient, edge curling coefficient, porosity expansion coefficient, cover reduction coefficient, and soil exposure coefficient. The formula is as follows:

[0121]

[0122] In the formula, This represents the mulch film state change index of the i-th mulch film monitoring unit during the t-th monitoring period; Indicates the color attenuation coefficient; Indicates the texture degradation coefficient; Indicates the edge curl factor; Indicates the hole expansion coefficient; Indicates the coverage reduction factor; Indicates the soil exposure coefficient; , , , , and The values ​​represent weighting coefficients, and a1+a2+a3+a4+a5+a6=1. Preferably, a1=0.14, a2=0.16, a3=0.18, a4=0.23, a5=0.14, and a6=0.15. Since the porosity expansion directly reflects the degree of transformation of the mulch film from continuous coverage to fragmented tearing, and edge curling can reflect the early structural instability caused by wind, soil moisture, and material softening, the porosity expansion coefficient and edge curling coefficient are set to relatively high weights.

[0123] Table 2: Examples of Calculation of Mulch Film State Change Index for Seven Mulch Film Monitoring Units

[0124] M01 0.46 0.44 0.40 0.36 0.38 0.30 0.388 M02 0.62 0.60 0.72 0.80 0.59 0.70 0.684 M03 0.22 0.21 0.18 0.12 0.24 0.10 0.174 M04 0.70 0.68 0.82 0.90 0.70 0.80 0.779 M05 0.50 0.49 0.45 0.43 0.43 0.38 0.448 M06 0.38 0.37 0.33 0.31 0.42 0.28 0.343 M07 0.55 0.54 0.60 0.69 0.55 0.62 0.597

[0125] As shown in Table 2, the mulch film state change index for M04 is 0.779, indicating that it exhibits large-area porosity, edge curling, and soil exposure at the pixel level. The mulch film state change index for M03 is 0.174, indicating a relatively low current visible degradation level; however, it needs to be considered in conjunction with the standard degradation process on day 28 to determine if there is any localized delayed degradation. The above calculations transform the complex pixel identification results into a unified index, facilitating subsequent temporal evolution judgment and input to the deep Q-network.

[0126] Example 3

[0127] This embodiment, based on Embodiment 2, further illustrates the specific implementation of the time-series evolution assessment module. This module includes a change amplitude calculation unit, a reference process reading unit, a time-series evolution coefficient calculation unit, and a degradation state determination unit. The system is set to monitor once every 7 days, with the 28th day after installation as the current monitoring period.

[0128] The change magnitude calculation unit calculates the increment of state change based on the mulch film state change index of the current monitoring period and the mulch film state change index of the previous monitoring period, using the following formula:

[0129]

[0130] In the formula, This represents the increment of the state change of the i-th mulch film monitoring unit during the t-th monitoring period; This index represents the change in the state of the plastic film during the current monitoring period. This represents the index of changes in the state of the plastic film during the previous monitoring period.

[0131] The reference process reading unit reads the standard degradation reference index D corresponding to the current monitoring cycle based on the type of mulch film material, the laying time, the crop growth stage, and field environmental data. ref (t), Standard degradation change increment ΔD ref (t), reference value for hole expansion H ref (t) and soil exposure reference value B ref (t). In this embodiment, the current period is the 28th day after laying, preferably, D ref (t)=0.52, ΔD ref (t)=0.10, H ref (t)=0.40, B ref (t)=0.35.

[0132] The temporal evolution coefficient calculation unit calculates the temporal evolution coefficient of the plastic film degradation, denoted as DTE, using the following formula:

[0133]

[0134] In the formula, This represents the temporal evolution coefficient of mulch film degradation for the i-th mulch film monitoring unit during the t-th monitoring period; Indicates the index of changes in the state of the plastic film; D ref (t) represents the standard degradation reference index corresponding to the current monitoring period; ΔD represents the increment of state change. ref (t) represents the standard degradation change increment; represents the pore expansion coefficient; Href(t) represents the pore expansion reference value corresponding to the current monitoring cycle; Indicates the soil exposure coefficient; B ref (t) represents the soil exposure reference value corresponding to the current monitoring period; , , and Let b1 represent the weighting coefficients, and b1+b2+b3+b4=1; ε represents a constant to prevent the denominator from being zero, with a preferred value of 0.000001. Preferably, b1=0.28, b2=0.27, b3=0.25, and b4=0.20.

[0135] The degradation state determination unit compares the degradation time-series evolution coefficient of the mulch film with a preset first time-series evolution threshold TDTE1 and a second time-series evolution threshold TDTE2. Preferably, TDTE1 = 0.60 and TDTE2 = 1.25.

[0136] When the time evolution coefficient of the mulch film degradation is greater than or equal to 0.60 and less than or equal to 1.25, the corresponding mulch film monitoring unit is determined to be in a normal degradation state.

[0137] When the time evolution coefficient of membrane degradation is greater than 1.25, and H i (t)≥TH、B i When (t)≥TB, the corresponding mulch film monitoring unit is determined to be in an abnormal premature rupture state; TH represents the abnormal hole expansion threshold, TB represents the abnormal soil exposure threshold, preferably, TH=0.65, TB=0.55.

[0138] When the degradation time evolution coefficient of the mulch film is less than 0.60, and the color attenuation coefficient, texture deterioration coefficient, and coverage reduction coefficient do not meet the standard degradation requirements for the corresponding monitoring period, the corresponding mulch film monitoring unit is determined to be in a state of local delayed degradation.

[0139] Table 3: Time-series evolution coefficients and state determination examples for seven mulch film monitoring units

[0140] M01 0.370 0.388 0.018 0.646 H=0.36; B=0.30 Normal degradation M02 0.420 0.684 0.264 1.949 H=0.80; B=0.70 Abnormal premature rupture M03 0.200 0.174 -0.026 0.168 H=0.12; B=0.10 Local delayed degradation M04 0.550 0.779 0.229 2.040 H=0.90; B=0.80 Abnormal premature rupture M05 0.400 0.448 0.048 0.832 H=0.43; B=0.38 Normal degradation M06 0.310 0.343 0.033 0.623 H=0.31; B=0.28 Normal degradation M07 0.360 0.597 0.237 1.720 H=0.69; B=0.62 Abnormal premature rupture

[0141] As shown in Table 3, M02, M04, and M07 exhibited simultaneous growth in porosity expansion and soil exposure during the early monitoring stages, and were therefore identified as abnormally premature ruptures. M03, however, did not show sufficient changes in color, texture, and cover reduction by day 28, and was therefore identified as a locally delayed degradation. This processing logic demonstrates that the present invention does not merely determine whether the current image is damaged, but rather compares the pixel-level change rate of the mulch film over multiple monitoring periods with a reference degradation process.

[0142] Example 4

[0143] This embodiment, based on embodiments one through three, further illustrates the specific implementation of the state space construction module and the deep Q-network decision module. The state space construction module includes a monitoring reliability coefficient calculation unit, an environmental correction coefficient calculation unit, a residual risk index calculation unit, and a state vector generation unit.

[0144] The monitoring reliability coefficient calculation unit calculates the monitoring reliability coefficient based on image clarity, crop occlusion ratio, soil occlusion ratio, and reflective interference ratio, using the following formula:

[0145]

[0146] In the formula, This represents the monitoring reliability coefficient of the i-th mulch film monitoring unit during the t-th monitoring period; The image sharpness coefficient is represented by Laplacian variance, edge sharpness, and motion blur estimation. The crop occlusion ratio is represented by the ratio of the number of crop occlusion pixels to the total number of pixels in the monitoring unit. The percentage of soil obstruction is indicated by the ratio of the number of pixels obstructed by soil to the number of pixels initially covered by the plastic film. The reflective interference ratio is indicated by the ratio of the number of reflective interference pixels to the total number of pixels in the monitoring unit. , , and Let c1 represent the weighting coefficients, and c1+c2+c3+c4=1. Preferably, c1=0.38, c2=0.26, c3=0.21, and c4=0.15.

[0147] It should be noted that, based on the first The monitoring unit for plastic film was in the first The image sharpness coefficient is calculated from the image data within a monitoring period and denoted as . Image sharpness coefficient Used to characterize whether the images within the current monitoring period are sufficient to support the identification of plastic film pixels, hole boundaries, edge curling, and soil exposure areas.

[0148] The system first reads the first The monitoring unit for plastic film was in the first The registered images were processed within a monitoring period, and pixels obscured by crops, soil, and reflective interference were excluded from the effective judgment area to obtain the effective pixel set for image sharpness assessment, denoted as [image set]. The number of pixels is denoted as The system records the grayscale image corresponding to this region as... ,in Indicates the pixel position.

[0149] The system processes grayscale images Perform Laplacian operator processing to obtain pixels. Laplace response value at And calculate the mean of the Laplacian response within the effective pixel set:

[0150]

[0151] In the formula, Indicates the first The monitoring unit for plastic film was in the first Mean Laplace response over a monitoring period; Represents the set of valid pixels in an image The number of pixels within; Represents pixels The Laplace response value at that location; Represents pixels It belongs to the set of valid pixels in the image.

[0152] Further calculation of Laplace detail sub-scores:

[0153]

[0154] In the formula, Indicates the first The monitoring unit for plastic film was in the first Laplace detail sub-scores within each monitoring period; The Laplace variance reference value is obtained through calibration using the initial sharp sampled image and manually confirmed sharp images. Represents pixels The Laplace response deviates from the square of the mean. The clearer the local membrane fiber texture, pore boundaries, and bare soil boundaries, the better. The closer .

[0155] The system in the effective pixel set of the image The edges of the plastic film, the boundaries of holes, the boundaries of cracks, and the boundaries of exposed soil are extracted to obtain a set of sharp edge pixels, denoted as [missing information]. The number of pixels is denoted as For any pixel within the set of pixels with sharpness edges. The system reads the horizontal gradient. and vertical gradient Calculate the edge sharpness sub-score:

[0156]

[0157] In the formula, Indicates the first The monitoring unit for plastic film was in the first Edge sharpness sub-score within each monitoring period; The clear edge gradient reference value is obtained by calibration using an initial clear sampled image, manually confirmed boundary images, and historical monitoring samples. Represents the set of pixels at the edge of sharpness; Indicates the number of pixels at the edge of sharpness; Represents pixels The horizontal gradient at that location; Represents pixels The vertical gradient at the location. When the transition between light and dark is more pronounced at the edge of the membrane paper, the boundary of the pores, and the boundary of the exposed soil, The closer .

[0158] The system uses the sharpness edge pixel set The estimated edge trailing width, unidirectional texture extension width, and edge transition width are the first parameters. The monitoring unit for plastic film was in the first The average trail width within a monitoring period is denoted as . And calculate the motion blur suppressor score:

[0159]

[0160] In the formula, Indicates the first The monitoring unit for plastic film was in the first Motion fuzziness suppressor score within each monitoring period; The average trailing width of the current monitoring unit image is estimated by edge trailing width, same-direction texture extension width, and edge transition width. This represents a reference value for the maximum permissible motion blur width, obtained through calibration using clear samples of drone images, mobile inspection images, and fixed camera images. The lighter the image motion blur, the better. The smaller, The closer When the image has obvious motion blur, reduce.

[0161] In obtaining Laplace detail sub-scores Edge sharpness sub-score and motion blur suppressor score Then, the system calculates the image sharpness coefficient according to the following formula. :

[0162]

[0163]

[0164] In the formula, Indicates the first The monitoring unit for plastic film was in the first Image clarity coefficient within a monitoring period; Indicates the Laplace sub-score; Indicates the edge sharpness sub-score; Indicates the motion blur suppressor score; Indicates the Laplace detail weights; Indicates the edge sharpness weight; This represents the motion blur suppression weights. Preferably, , , The sum of the above weights is Since the identification of plastic film degradation mainly relies on the paper fiber texture, crack boundaries, hole boundaries, and bare soil exposure boundaries, Laplacian detail and edge sharpness have a significant impact on the identification results; motion blur is mainly used to determine whether there is directional ghosting in the current image, and therefore serves as an auxiliary constraint on image sharpness.

[0165] The environmental correction factor calculation unit calculates the environmental correction factor based on soil moisture, rainfall, air temperature, and wind speed, using the following formula:

[0166]

[0167] In the formula, W represents the environmental correction factor for the i-th mulch film monitoring unit during the t-th monitoring period; i (t) represents the normalized soil moisture effect value; This represents the normalized impact value of rainfall; This represents the normalized temperature effect value; The value represents the normalized wind speed influence; g1, g2, g3 and g4 represent environmental correction weights, with preferred values ​​of 0.25, 0.30, 0.20 and 0.25, respectively.

[0168] The residual risk index calculation unit calculates the residual risk index based on the density of plastic film fragments, the residual coverage ratio, the spatial connectivity of the fragments, and the degree of delayed degradation, using the following formula:

[0169]

[0170] In the formula, This represents the residual risk index of the i-th mulch film monitoring unit during the t-th monitoring period; The density of plastic film fragments is indicated by the number of broken fragments per unit area. The residual coverage ratio is indicated by the ratio of the area of ​​residual plastic film to the area of ​​the monitoring unit. The degree of spatial connectivity of fragments is represented by the distance between the centers of adjacent fragments and the number of connected edges. The degree of delayed degradation is indicated by the difference between the standard degradation reference index and the actual mulch film state change index. The standard degradation reference index is derived from data on the type of mulch film material, laying time, crop growth stage, and field environment. In other words, the same type of mulch film should exhibit different degrees of degradation on days 7, 14, 21, and 28 after laying. At the same laying time, different soil moisture, rainfall, temperature, and wind speed will also affect the normal degradation rate. The standard degradation reference index is compared with the actual mulch film state change index. If the actual mulch film state change index has reached the standard degradation reference requirement for the current period, the degradation progress of the monitoring unit is normal and not considered delayed degradation. If the actual mulch film state change index has not reached the standard degradation reference requirement for the current period, it indicates insufficient changes in the color, texture, coverage reduction, or damage of the mulch film in that area, and the system marks it as a candidate area for delayed degradation. To avoid misjudgment, the system further examines specific degradation characteristics. If the actual mulch film state change index is low, and the color decay, texture deterioration, and coverage reduction are not significant, and the porosity expansion and soil exposure do not reach the levels expected for the current cycle, then it indicates that there is indeed an insufficient degradation problem in this area, and its degree of delayed degradation can be confirmed. Conversely, if the low actual mulch film state change index is caused by crop shading, reflection, soil cover, or image blurring, the system will not directly identify it as delayed degradation, but will trigger the fusion of adjacent cycles or resampling based on the monitoring reliability coefficient. Confirmation is also based on continuous monitoring cycles. If a monitoring unit shows insufficient degradation only in one cycle, but returns to normal in subsequent cycles, the system will not consider it a stable delayed degradation area; if the monitoring unit is below the standard degradation reference requirement for the corresponding cycle in multiple consecutive monitoring cycles, and the color, texture, and coverage status continue to fail to meet the normal degradation requirements, then it is confirmed as a localized delayed degradation. The greater the difference between the standard degradation reference index and the actual mulch film state change index, and the more consecutive cycles, the more likely it is to be considered delayed degradation. The higher the value, the better. d1, d2, d3, and d4 represent weighting coefficients, and d1+d2+d3+d4=1. Preferably, d1=0.32, d2=0.25, d3=0.23, and d4=0.20.

[0171] The state vector generation unit combines the mulch film state change index, mulch film degradation time-series evolution coefficient, monitoring reliability coefficient, environmental correction coefficient, residual risk index, state change increment, crop shading ratio, and reflective interference ratio to generate a deep Q-network state vector, as shown in the following formula:

[0172]

[0173] In the formula, Let represent the state vector of the deep Q network for the i-th monitoring unit of the plastic film during the t-th monitoring period. This state vector is jointly input into the deep Q network with the image pixel recognition results, degradation time-series evolution results, environmental correction results, and monitoring reliability results, enabling the deep Q network to distinguish between "real breakage", "low-reliability images", "suspected damage caused by occlusion", and "residual risk caused by delayed degradation".

[0174] Action space construction unit constructs a set of actions for processing monitoring data of plastic film, including Normal degradation tracking action, Abnormal early rupture warning action, Local delayed degradation confirmation action, Residual risk marking action, Adjacent cycle fusion actions, Local image enhancement actions and The sampling action is reviewed. The action value index calculation unit inputs the state vector of the deep Q-network into the deep Q-network and calculates the action value index using the following formula:

[0175]

[0176] In the formula, This represents the action value index of the k-th mulch film monitoring data processing action within the t-th monitoring period; This represents the state vector of a deep Q-network; θ represents the data processing action for the kth monitoring data of the plastic film; θ represents the depth Q network parameters.

[0177] The action selection unit sorts the action value indices and selects the action with the highest value index as the target action, as shown in the following formula:

[0178]

[0179] In the formula, This represents the target processing action of the i-th mulch film monitoring unit during the t-th monitoring period; This indicates that the action with the highest action value index is selected from all candidate actions. When the difference between the highest and second-highest action value index is less than the action value difference threshold TQ, the system adjusts the target processing action to a review sampling action. Preferably, TQ = 0.06.

[0180] Table 4: Examples of State Space and Motion Value Decision-Making for Seven Mulch Film Monitoring Units

[0181] M01 0.388 0.646 0.894 0.149 A1 0.82 Regular tracking M02 0.684 1.949 0.865 0.307 A2 0.88 Abnormal early rupture warning M03 0.174 0.168 0.914 0.218 A3 0.84 Confirmation of localized delayed degradation M04 0.779 2.040 0.813 0.601 A4 0.91 Residual risk marking M05 0.448 0.832 0.742 0.199 A6 0.76 Local image enhancement M06 0.343 0.623 0.684 0.201 A7 0.79 Verification sampling M07 0.597 1.720 0.843 0.421 A2 0.86 Abnormal early rupture warning

[0182] As shown in Table 4, although M04 exhibits abnormal rupture, its residual risk index has reached 0.601. Therefore, the deep Q network assigns the highest action value to the A4 residual risk marker. The monitoring confidence coefficient of M06 is 0.684, which is lower than the low confidence threshold of 0.70. Therefore, the system does not directly output a degradation conclusion but generates a verification sampling task. This processing method can avoid misjudgments caused by crop shading, soil cover, and reflective interference.

[0183] Example 5

[0184] This embodiment, based on Embodiment 4, further illustrates the specific implementation of the deep Q-network decision-making module and the processing result output module. The deep Q-network decision-making module includes a reward weight configuration unit, a reward update coefficient calculation unit, a target action value calculation unit, a network loss calculation unit, and a network parameter update unit; the processing result output module includes a partitioning result generation unit, a risk level determination unit, and a monitoring task update unit. The reward weight configuration unit is used to configure the weights corresponding to the manual review consistency value, subsequent periodic consistency value, residual risk hit value, review sampling cost, and misjudgment penalty value; the reward update coefficient calculation unit is used to calculate the reward update coefficient; the target action value calculation unit is used to calculate the target action value of the deep Q-network; the network loss calculation unit is used to calculate the difference between the current action value and the target action value; the network parameter update unit is used to update the deep Q-network parameters; the partitioning result generation unit is used to generate farmland spatial partitioning results; the risk level determination unit is used to output the risk level based on the residual risk index; and the monitoring task update unit is used to output subsequent monitoring tasks based on the risk level and the target processing action.

[0185] The reward update coefficient calculation unit calculates the reward update coefficient based on the consistency value of manual review, the consistency value of subsequent cycles, the residual risk hit value, the review sampling cost, and the misjudgment penalty value, as shown in the following formula:

[0186]

[0187] In the formula, This represents the reward update coefficient for the i-th mulch film monitoring unit during the t-th monitoring period; This indicates a value that has been manually verified to be consistent. This represents the consistency value for subsequent cycles, used to characterize the degree of consistency between the current judgment and the evolution results of subsequent monitoring cycles; Indicates the residual risk hit value; Indicates the cost of verifying the sampling; Indicates the penalty value for misjudgment; , , , and These represent weighting coefficients. Preferably, e1=0.34, e2=0.25, e3=0.20, e4=0.09, and e5=0.12.

[0188] The network parameter update unit updates the parameters of the depth-Q network using the target action value update method, where the target action value is denoted as... The formula is as follows:

[0189]

[0190] In the formula, γ represents the target action value of the i-th mulch film monitoring unit in the t-th monitoring period; γ represents the discount factor, with a preferred value of 0.90; This represents the state vector of the deep Q-network for the next monitoring period; This indicates the processing action for the k-th monitoring data of the plastic film mulch. This represents the target network parameters.

[0191] The network parameter update unit further constructs a loss function based on the difference between the target action value and the current action value, as shown in the following formula:

[0192]

[0193] In the formula, This represents the depth Q-network update loss of the i-th mulch film monitoring unit during the t-th monitoring period; θ represents the selected target processing action within the current monitoring period; θ represents the online network parameters. Through the above reward update and loss function calculation, the deep Q-network can continuously learn the action selection rules under different stages of mulch film degradation, different pixel interference types, and different residual risk states.

[0194] The risk level determination unit compares the residual risk index with a first residual risk threshold TRRI1 and a second residual risk threshold TRRI2. Preferably, TRRI1 = 0.25 and TRRI2 = 0.55.

[0195] When the residual risk index is less than 0.25, the first residual risk level is generated; when the residual risk index is greater than or equal to 0.25 and less than 0.55, the second residual risk level is generated; when the residual risk index is greater than or equal to 0.55, the third residual risk level is generated.

[0196] Table 5: Examples of Reward Updates and Task Outputs for the Seven Mulch Film Monitoring Units

[0197] M01 A1 Normal Tracking 0.149 First level 0.10 0.00 0.586 Routine periodic monitoring M02 A2 rupture warning 0.307 Second level 0.25 0.00 0.770 Encrypted monitoring + partial verification M03 A3 Delayed Confirmation 0.218 First level 0.20 0.05 0.514 Routine periodic monitoring M04 A4 Risk Marking 0.601 Level 3 0.35 0.00 0.759 Residue warning + Recycling operation warning M05 A6 Image Enhancement 0.199 First level 0.30 0.10 0.397 Routine periodic monitoring M06 A7 Re-examination Sampling 0.201 First level 0.45 0.00 0.449 Routine periodic monitoring M07 A2 rupture warning 0.421 Second level 0.30 0.05 0.697 Encrypted monitoring + partial verification

[0198] As shown in Table 5, the residual risk index of M04 is 0.601, reaching the third residual risk level, and the system outputs residual risk warnings and reminders for the recycling of plastic film. Although M02 and M07 mainly exhibit abnormal premature rupture, their residual risk indices are at the second residual risk level, so the system outputs encrypted monitoring and local verification sampling tasks. M06 triggers verification sampling due to its low monitoring confidence coefficient, which can provide effective training samples related to low-confidence images for the deep Q network.

[0199] This invention, through the above embodiments, achieves pixel-level recognition, exponential expression, temporal evolution judgment, and deep Q-network action selection for changes in the state of biodegradable plastic film. The system can distinguish between normal degradation, abnormal premature breakage, locally delayed degradation, and residual risk within the same technical framework, avoiding misjudgments caused by relying solely on single image damage detection. This makes the plastic film monitoring data processing continuous, adaptive, and trainable / updable.

[0200] Furthermore, in the Deep Q-Network, Q represents the action value function, which characterizes the long-term processing benefit obtained after performing a certain data processing action under the current monitoring state of the plastic film. In this embodiment, the Deep Q-Network does not directly classify plastic film images. Instead, it receives the Deep Q-Network state vector output by the state space construction module and learns the data processing action selection rules under different plastic film degradation states, different image interference conditions, and different residual risk states. Specifically, the Deep Q-Network model first determines the state space, action space, reward feedback, and network parameter update method. The state space consists of the plastic film state change index, plastic film degradation temporal evolution coefficient, monitoring reliability coefficient, environmental correction coefficient, residual risk index, state change increment, crop shading ratio, and reflective interference ratio of the i-th plastic film monitoring unit in the t-th monitoring period. These state quantities can simultaneously characterize the current degree of damage to the plastic film, temporal evolution speed, image usability, field environmental impact, and residual risk level, enabling the Deep Q-Network to distinguish between actual breakage, low-reliability images, suspected damage caused by shading, and residual risk caused by delayed degradation. The action space includes normal degradation tracking actions, abnormal early breakage warning actions, local delayed degradation confirmation actions, residual risk marking actions, adjacent period fusion actions, local image enhancement actions, and verification sampling actions. The deep Q-network takes a state vector as input and outputs the action value index corresponding to each candidate data processing action. A higher action value index indicates that performing the action under the current monitoring state is more conducive to obtaining accurate degradation judgment results and residual risk identification results.

[0201] During the training of the deep Q-network model, the system uses state vectors from historical monitoring periods, actual actions performed, manual review results, evolution results of subsequent periods, residual risk hit results, review sampling costs, and misjudgment penalty results to form training samples. The reward update coefficient calculation unit calculates the reward update coefficient based on the consistency value of manual review, the consistency value of subsequent periods, the residual risk hit value, the review sampling cost, and the misjudgment penalty value. If the current action is consistent with the manual review result, and the subsequent monitoring period verifies that the action judgment is correct, the corresponding action value is increased; if the current action leads to a misjudgment of premature anomaly rupture, a missed judgment of local delayed degradation, or an incorrect labeling of residual risk, the corresponding action value is decreased. Furthermore, the network parameter update unit uses an online network and a target network for training. The online network is used to output the action value index of each data processing action in the current state, and the target network is used to calculate the target action value in the next monitoring period. The system constructs the target action value based on the reward update coefficient, the discount factor, and the maximum action value in the next period, calculates the network loss through the difference between the target action value and the current action value, and then updates the online network parameters based on the network loss. After multiple rounds of training with monitoring samples, the deep Q network can adaptively select normal tracking, early warning, confirmation, marking, fusion, enhancement or verification sampling actions under different stages of mulch film degradation, different shading and reflection conditions and different residual risk levels, thereby improving the stability and continuity of mulch film monitoring data processing results.

[0202] It should be noted that all calculation formulas in this application employ regression analysis, including but not limited to machine learning algorithms, to deeply analyze the collected parameters and identify their natural trends and interrelationships. Specialized software, such as Python's Scikit-learn library or the R language, is used to automatically generate mathematical models that match the data. Then, cross-validation and other methods are used to objectively evaluate the model performance, and continuous feedback and optimization are combined to ensure that the created formulas truly reflect the inherent laws of the data, thereby guaranteeing their effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula undergo dimensionless processing within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless processing techniques include, but are not limited to, min-max-normalization and Z-score standardization.

[0203] The algorithm of this invention is implemented as a Python script. Before executing the core logic, the program first executes a data loading module (e.g., using the widely used pandas library in Python) configured to read the aforementioned spreadsheet file and load its contents into the program's working memory (e.g., a DataFrame data structure). Subsequent algorithm steps will directly query and retrieve the required configuration parameters from this in-memory data structure.

[0204] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A data processing system for monitoring agricultural mulch film based on a deep Q-network, applied to monitoring farmland covered with biodegradable agricultural mulch film, characterized in that, include: The monitoring data acquisition module is used to divide the mulch film monitoring units and collect mulch film image data, spatial location data, collection time data and field environment data within multiple monitoring periods to form a multi-period mulch film monitoring dataset. The degradation feature calculation module is used to calculate the pixel likelihood value of the mulch film based on the multi-period mulch film monitoring dataset, and combine it with the interference mask, the coverage state mask, the soil exposure pixel mask and the hole candidate pixel mask to obtain the color attenuation coefficient, texture degradation coefficient, edge curling coefficient, hole expansion coefficient, coverage reduction coefficient and soil exposure coefficient, and then generate the mulch film status change index. The temporal evolution assessment module is used to generate the temporal evolution coefficient of mulch film degradation based on the mulch film state change index, state change increment, pore expansion coefficient and soil exposure coefficient, and to determine the degradation state. The state space construction module is used to generate a deep Q-network state vector based on the mulch film state change index, mulch film degradation time-series evolution coefficient, monitoring reliability coefficient, environmental correction coefficient, and residual risk index. The deep Q-network decision module is used to output the action value index corresponding to the data processing action based on the state vector of the deep Q-network, and to determine the target processing action. The processing result output module is used to output degradation partitioning results and monitoring tasks based on the target processing actions.

2. The data processing system for monitoring plastic film based on deep Q-networks according to claim 1, characterized in that: The monitoring data acquisition module includes a segmentation unit, an image data acquisition unit, an environmental data acquisition unit, and a periodic registration unit; The division unit is used to divide the farmland monitoring area into several mulch film monitoring units based on the mulch film laying boundary, crop planting row spacing, planting ridge direction and field boundary, and to generate a unit number for each mulch film monitoring unit. The image data acquisition unit is used to acquire top-view images of the ridge surface using a UAV multispectral camera, acquire images of the edge of the mulch film using a ground-fixed camera, and acquire images of the hole area and the soil exposed area using a mobile inspection camera. The environmental data acquisition unit is used to collect data on light intensity, soil moisture, rainfall, temperature and wind speed through field environmental sensors to form field environmental data. The periodic registration unit is used to spatially register mulch film image data collected in different monitoring periods based on the intersection of ridge lines, the corner points of field boundaries, and the initial laying outline of mulch film. The registered mulch film image data is then associated with the collection time data, unit number, and field environment data to form a continuous monitoring sequence for the same mulch film monitoring unit.

3. The data processing system for monitoring plastic film based on deep Q-networks according to claim 1, characterized in that: The degradation feature calculation module includes a pixel-level feature extraction unit, a mulch film pixel likelihood calculation unit, an interference mask generation unit, a coverage state mask generation unit, a hole connectivity identification unit, and a basic degradation feature calculation unit. The pixel-level feature extraction unit is used to extract the brightness similarity score, color similarity score, near-infrared discrimination score, edge continuity score, and paper fiber texture score of the current pixel based on the Lab color space, HSV color space, near-infrared channel, edge detection results, and local texture window. The mulch film pixel likelihood calculation unit is used to fuse the brightness similarity score, chromaticity similarity score, near-infrared differentiation score, edge continuity score and paper fiber texture score to obtain the pixel likelihood value of the current pixel belonging to the mulch film. The interference mask generation unit is used to generate a crop occlusion pixel mask based on the vegetation index, a reflective interference pixel mask based on the HSV spatial brightness and saturation values, and a soil adhesion pixel mask based on the soil color likelihood value, the initial mulch film covering mask, and the neighboring mulch film support ratio; the crop occlusion pixel mask, the reflective interference pixel mask, and the soil adhesion pixel mask together constitute the interference mask. The coverage state mask generation unit is used to generate a clean mulch paper effective pixel mask, a coverage presence mask, and an effective determination area mask based on the mulch paper pixel likelihood value, the initial mulch paper coverage mask, the crop shading pixel mask, the soil adhesion pixel mask, and the reflective interference pixel mask. The hole connected component identification unit is used to generate a hole candidate pixel mask based on the initial mulch film covering mask, the effective determination area mask, the covering existence mask, the soil exposed pixel mask, and the hole pixel likelihood value, and to perform morphological processing and connected component labeling on the hole candidate pixel mask to obtain a set of hole connected components. The basic degradation feature calculation unit is used to calculate the color attenuation coefficient, texture degradation coefficient, edge curling coefficient, hole expansion coefficient, coverage reduction coefficient, and soil exposure coefficient based on the effective pixel mask of clean mulch film, the cover existence mask, the effective determination area mask, the soil exposure pixel mask, the set of hole connected domains, and the current edge pixel set of mulch film.

4. The data processing system for monitoring plastic film based on deep Q-networks according to claim 3, characterized in that: The degradation feature calculation module is also used to generate a mulch film state change index based on color attenuation coefficient, texture deterioration coefficient, edge curling coefficient, hole expansion coefficient, coverage reduction coefficient, and soil exposure coefficient. Among them, the color attenuation coefficient is used to characterize the degree of color change of the effective pixels of the clean mulch film relative to the initial mulch film; the texture degradation coefficient is used to characterize the degree of change of paper fiber texture and crack line density; the edge curling coefficient is used to characterize the degree of offset of the current edge relative to the initial reference edge line; the hole expansion coefficient is used to characterize the number of holes, the hole area and the hole area growth rate; the coverage reduction coefficient is used to characterize the degree of reduction of the proportion of mulch film coverage in the effective determination area; and the soil exposure coefficient is used to characterize the proportion of bare soil exposure within the initial coverage area.

5. The data processing system for monitoring mulch film based on a deep Q-network according to claim 4, characterized in that: The temporal evolution evaluation module includes a change amplitude calculation unit, a reference process reading unit, a temporal evolution coefficient calculation unit, and a degradation state determination unit; The change amplitude calculation unit is used to calculate the state change increment based on the mulch film state change index of the current monitoring period and the mulch film state change index of the previous monitoring period. The reference process reading unit is used to read the standard degradation reference index, standard degradation change increment, pore expansion reference value and soil exposure reference value corresponding to the current monitoring cycle based on the type of mulch film material, laying time, crop growth stage and field environment data. The time-series evolution coefficient calculation unit is used to calculate the time-series evolution coefficient of mulch film degradation based on the correspondence between the mulch film state change index and the standard degradation reference index, the correspondence between the state change increment and the standard degradation change increment, the correspondence between the pore expansion coefficient and the pore expansion reference value, and the correspondence between the soil exposure coefficient and the soil exposure reference value. The degradation state determination unit is used to compare the degradation time evolution coefficient of the mulch film with the first time evolution threshold and the second time evolution threshold, and combine the pore expansion coefficient, soil exposure coefficient, color attenuation coefficient, texture deterioration coefficient and coverage reduction coefficient to output the degradation state corresponding to the mulch film monitoring unit.

6. The data processing system for monitoring mulch film based on a deep Q-network according to claim 5, characterized in that: The degradation state determination unit is further used for: When the time-series evolution coefficient of the mulch film degradation is greater than or equal to the first time-series evolution threshold and less than or equal to the second time-series evolution threshold, the corresponding mulch film monitoring unit is determined to be in a normal degradation state. When the degradation time-series evolution coefficient of the mulch film is greater than the second time-series evolution threshold, and the pore expansion coefficient is greater than or equal to the abnormal pore expansion threshold and the soil exposure coefficient is greater than or equal to the abnormal soil exposure threshold, the corresponding mulch film monitoring unit is determined to be in an abnormal premature rupture state. When the degradation time-series evolution coefficient of the mulch film is less than the first time-series evolution threshold, and the color decay coefficient, texture deterioration coefficient, and coverage reduction coefficient do not meet the standard degradation requirements for the corresponding monitoring period, the corresponding mulch film monitoring unit is determined to be in a state of local delayed degradation.

7. The data processing system for monitoring mulch film based on a deep Q-network as described in claim 1, characterized in that: The state space construction module includes a monitoring reliability coefficient calculation unit, an environmental correction coefficient calculation unit, a residual risk index calculation unit, and a state vector generation unit; The monitoring reliability coefficient calculation unit is used to calculate the monitoring reliability coefficient based on image clarity, crop occlusion ratio, soil occlusion ratio, and reflective interference ratio. The environmental correction coefficient calculation unit is used to calculate the environmental correction coefficient based on soil moisture, rainfall, air temperature and wind speed; The residual risk index calculation unit is used to calculate the residual risk index based on the density of plastic film paper fragments, the residual coverage ratio, the spatial connectivity of the fragments, and the degree of delayed degradation. The state vector generation unit is used to combine the mulch film state change index, mulch film degradation time evolution coefficient, monitoring reliability coefficient, environmental correction coefficient, residual risk index, state change increment, crop shading ratio, and reflective interference ratio into a deep Q network state vector.

8. The data processing system for monitoring mulch film based on a deep Q-network according to claim 7, characterized in that: The deep Q-network decision module includes an action space construction unit, an action value index calculation unit, and an action selection processing unit. The action space construction unit is used to construct a set of actions for processing monitoring data of plastic film, which includes normal degradation tracking action, abnormal early breakage warning action, local delayed degradation confirmation action, residual risk marking action, adjacent period fusion action, local image enhancement action, and verification sampling action. The action value index calculation unit is used to input the state vector of the deep Q network into the deep Q network and calculate the action value index corresponding to each action of the mulch film monitoring data processing action. The processing action selection unit is used to sort the value indices of each action and take the processing action of the mulch film monitoring data with the highest action value index as the target processing action of the current mulch film monitoring unit; when the difference between the highest action value index and the second highest action value index is less than the action value difference threshold, the target processing action is adjusted to the verification sampling action.

9. The data processing system for monitoring mulch film based on a deep Q-network as described in claim 8, characterized in that: The deep Q-network decision module also includes a reward weight configuration unit, a reward update coefficient calculation unit, a target action value calculation unit, a network loss calculation unit, and a network parameter update unit; The reward weight configuration unit is used to configure the weights corresponding to the consistency value of manual review, the consistency value of subsequent cycles, the residual risk hit value, the review sampling cost, and the misjudgment penalty value. The reward update coefficient calculation unit is used to calculate the reward update coefficient based on the consistency value of manual review, the consistency value of subsequent cycles, the residual risk hit value, the review sampling cost, and the misjudgment penalty value. The target action value calculation unit is used to calculate the target action value based on the reward update coefficient, the deep Q network state vector of the next monitoring period, the target network parameters, and the discount factor. The network loss calculation unit is used to generate network update loss based on the difference between the target action value and the current action value; The network parameter update unit is used to update the parameters of the deep Q network according to the network update loss, so that the deep Q network can learn the action selection rules under different degradation stages, different pixel interference types and different residual risk states.

10. The data processing system for monitoring mulch film based on a deep Q-network according to claim 9, characterized in that: The processing result output module includes a partition result generation unit, a risk level determination unit, and a monitoring task update unit; The partitioning result generation unit is used to map the mulch film state change index, mulch film degradation time evolution coefficient, monitoring reliability coefficient, residual risk index and target treatment action corresponding to each mulch film monitoring unit to the farmland spatial layer, and generate normal degradation area, abnormal premature rupture area, locally delayed degradation area, low reliability verification area and residual risk area. The risk level determination unit is used to compare the residual risk index with the first residual risk threshold and the second residual risk threshold, and generate the first residual risk level, the second residual risk level or the third residual risk level. When the residual risk index reaches the residual risk determination condition, the corresponding mulch film monitoring unit is marked as having a residual risk status. When the residual risk index reaches the determination condition of the third residual risk level, the corresponding mulch film monitoring unit is marked as having a residual risk status. The monitoring task update unit is used to output regular periodic monitoring tasks, encrypted monitoring cycles and local verification sampling tasks, key area verification sampling tasks, residual risk warning tasks, mulch film recycling operation warning tasks, verification sampling actions or adjacent cycle fusion tasks based on risk level judgment results, monitoring reliability coefficient and target processing actions.

Images