Machine vision-based vegetable maturity online identification system for organic fertilizer production

Abstract

The invention relates to the technical field of organic fertilizer production, in particular to a machine vision-based vegetable decomposition degree online identification system for organic fertilizer production, which comprises a pile body image acquisition module, a permeation path construction module, a diffusion direction analysis module, a diffusion characteristic quantification module and a decomposition degree judgment module, the pile body image acquisition module is used for acquiring a surface image of a vegetable raw material pile body in an organic fertilizer composting area according to a preset time monitoring period and generating a timestamp index corresponding to an acquisition moment; according to the method, a decomposition permeation path chain is constructed, the physical process that vegetable decomposition liquid flows downwards along a pile gap under the traction of gravity is simulated, discrete dark spots are connected into link lines with physical significance, and the system can distinguish the decomposition depth by counting the number and length of the link lines, so that the decomposition efficiency is improved. Therefore, the identification result is more in line with the real biological reaction state in the vegetable heap.

Description

Technical Field

[0001] This invention relates to the field of organic fertilizer production technology, specifically to an online identification system for the maturity of vegetables used in organic fertilizer production based on machine vision. Background Technology

[0002] In the large-scale production of organic fertilizer, the composting and maturation of vegetable waste is a complex biochemical reaction process. Accurately judging the degree of maturation of vegetable raw materials is crucial for controlling the timing of turning the compost, ensuring fertilizer quality, and preventing the uncontrolled spread of decay pathogens.

[0003] In the actual production of organic fertilizer, vegetable raw material piles are usually made up of a large number of irregularly stacked leafy vegetables and stems, with a complex internal gap structure. Existing monitoring methods mainly rely on manual sensory or physicochemical testing with inserts. Not only are manual inspections difficult and pose safety hazards due to the presence of malodorous gases such as ammonia and hydrogen sulfide on site, but single-point sampling also makes it difficult to detect localized anaerobic decay inside the pile. Although some production lines have introduced basic machine vision technology, it only calculates the proportion of discolored areas on the surface to determine the degree of decay.

[0004] However, the nature of vegetable decay is often accompanied by tissue liquefaction. The decay liquid, influenced by gravity, seeps downwards along the stacking gaps, forming linear or chain-like traces with specific directions, rather than simply surface patches of color. Existing technologies cannot identify this gap flow characteristic that represents deep penetration, making it easy to misjudge shallow spots with dry surfaces as deep decay, or to overlook tiny but rapidly penetrating decay channels. In addition, vegetable decay is a dynamic and rapidly changing biological process. Existing systems mostly rely on single-photo assessments, making it impossible to determine whether a decaying area is stagnant or rapidly expanding. This makes it difficult to accurately guide the turning machine to intervene in the early stages of decay, and it is difficult to ensure the uniformity of fertilizer fermentation quality. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides the following technical solution: an online identification system for the composting degree of vegetables used in organic fertilizer production based on machine vision, comprising: The pile image acquisition module is used to acquire surface images of vegetable raw material piles in the organic fertilizer composting area according to a preset time monitoring cycle, and generate timestamp indexes corresponding to the acquisition time; locate rotten areas based on the color differences of the surface images, and extract the distribution data of mature dark spots associated with the timestamp index; The infiltration path construction module is used to analyze the distribution data of the decomposed dark spots, connect the spatially adjacent pixels one by one, simulate the flow traces of the decomposed liquid in the gaps of the pile body, and construct the decomposed infiltration path chain. The diffusion direction analysis module is used to analyze the extension trend of each of the aforementioned decay and penetration path chains, and select the direction with the highest proportion as the dominant decay diffusion direction. The diffusion feature quantification module is used to count the number and average length of the decomposition and infiltration path chains, and calculate the product of the two as the total decomposition feature; it generates a longitudinal infiltration enhancement marker based on the increase in the length of the decomposition and infiltration path chain, and generates a lateral diffusion addition marker based on the increase in the number of the decomposition and infiltration path chains. The decay maturity determination module is used to combine the dominant decay diffusion direction, the total decay characteristics, the longitudinal penetration enhancement marker, and the lateral diffusion addition marker to evaluate the structured spread state of the decay area on the surface of the pile body and generate the pile body decay maturity identification result.

[0006] Preferably, the decayed areas are located based on the color differences in the surface image, and the distribution data of decayed dark spots associated with the timestamp index is extracted, including: The surface image is converted into the HSV color space, the luminance component and the brightness component are separated, and single-channel grayscale data representing the depth of decay color is extracted. Set a decay grayscale segmentation threshold, perform binarization processing on the single-channel grayscale data, and mark pixels whose grayscale values ​​meet the preset decay conditions as dark spot candidate pixels. A morphological opening operation is performed on the region composed of the dark spot candidate pixels to eliminate discrete isolated noise and smooth the region edges, thereby generating a dark spot binarization mask. Traverse the binary mask of dark spots, extract the pixel coordinate indices of all mask regions, and associate and store the pixel coordinate indices with the timestamp index to generate the decayed dark spot distribution data.

[0007] Preferably, setting a grayscale segmentation threshold for ripening includes: The frequency of pixel occurrence at each gray level in the single-channel grayscale data is statistically analyzed, and a grayscale frequency histogram is constructed. Gaussian smoothing is applied to the grayscale frequency histogram to eliminate local noise fluctuations and extract the smoothed peak feature points; Identify the background wave peaks and the decay target wave peaks among the wave peak feature points, and locate the minimum point of the valley between the background wave peaks and the decay target wave peaks; Read the gray value corresponding to the minimum point of the trough and assign it as the value of the mature gray value segmentation threshold for subsequent processing.

[0008] Preferably, the distribution data of the decomposition dark spots is analyzed, and spatially adjacent pixels are connected step by step to simulate the flow traces of the decomposition liquid in the gaps of the pile body, constructing a decomposition and infiltration path chain, including: Extract the two-dimensional position coordinates of all dark spot pixels in the decayed dark spot distribution data, and establish a set of pixel coordinates to be traversed; Select any point from the set of pixel coordinates to be traversed as the current traversal node, and retrieve all discrete points within the preset neighborhood radius of the current traversal node to form a neighborhood candidate pixel group. Calculate the spatial Euclidean distance between the current traversed node and each pixel in the neighboring candidate pixel group, and lock the pixel with the smallest Euclidean distance as the secondary extension node; Establish a directional connection vector from the current traversed node to the secondary extended node, update the secondary extended node to the new current traversed node, and remove the connected pixels from the set of pixel coordinates to be traversed. The above retrieval, calculation, and locking steps are repeated until the updated current traversed node has no neighboring candidate pixels within the preset neighborhood radius. The generated series of directional connection vectors connected end to end are combined to obtain the decay and penetration path chain.

[0009] Preferably, the extension trend of each of the aforementioned decay and infiltration pathway chains is analyzed, and the direction with the highest proportion is selected as the dominant decay diffusion direction, including: Obtain the starting and ending pixel coordinates of each of the aforementioned decay and penetration path chains, and construct a penetration trend vector pointing from the starting pixel coordinates to the ending pixel coordinates; Calculate the angle between the penetration trend vector and the preset horizontal reference axis to obtain the directional angle characteristics of each path chain; Several consecutive angular statistical intervals covering 0 to 360 degrees are preset, and the directional angular features of all the path chains are mapped to the corresponding angular statistical intervals. The distribution frequency of the path chains contained in each angular statistical interval is counted. By numerically comparing the distribution frequencies of all the aforementioned path chains, the dominant angle interval corresponding to the maximum frequency value is identified. Extract the center angle value of the dominant angle range and determine it as the dominant diffusion direction of decay.

[0010] Preferably, the number and average length of the decomposition and infiltration pathway chains are statistically analyzed, and their product is calculated as the total decomposition characteristic, including: The total number of the aforementioned decay and infiltration pathway chains is counted and used as a characteristic of the number of links to characterize the dispersion of decay areas; Based on each of the aforementioned decomposition and infiltration path chains, extract all the directional connection vectors contained therein, and calculate the single-chain physical extension length of each path chain by accumulating Euclidean distances; Based on the link quantity characteristic, the arithmetic mean of the physical extension length of all the single links is calculated to obtain the average link length characterizing the decay depth. The link quantity feature is associated with the average link length and stored using the timestamp index. Then, a multiplication operation is performed to multiply the link quantity feature by the average link length to obtain the total amount of decayed link feature.

[0011] Preferably, generating longitudinal permeability enhancement markers based on the increase in the length of the decay permeation path chain includes: Obtain the average link length calculated for the current monitoring period, and obtain the historical average link length calculated for the previous monitoring period; The average link length of the current monitoring period is compared with the historical average link length. When the current value is greater than the historical value, a vertical penetration enhancement marker is generated.

[0012] Preferably, generating new lateral diffusion markers based on the increase in the number of the decomposition and infiltration pathway chains includes: Obtain the link quantity characteristics obtained from the statistics of the current monitoring period, and obtain the historical link quantity characteristics obtained from the statistics of the previous monitoring period; The link quantity characteristics of the current monitoring period are compared with the historical link quantity characteristics. When the current value is greater than the historical value, a new horizontal diffusion marker is generated.

[0013] Preferably, by combining the dominant decay diffusion direction, the total decay characteristics, the longitudinal penetration enhancement markers, and the lateral diffusion addition markers, the structured spread state of the decay area on the pile surface is evaluated, and a pile maturity identification result is generated, including: A multidimensional mathematical model for assessing decomposition is constructed, and the total decomposition characteristics are used as the benchmark assessment values ​​to characterize the current amount of decay. Calculate the cosine of the angle between the dominant decay diffusion direction and the preset vertical gravity axis to generate a direction weighting coefficient that reflects the vertical penetration ability of the decay liquid under the action of gravity. The presence status of the longitudinal penetration enhancement marker and the lateral diffusion addition marker is detected, and growth rate analysis is performed based on the detection results to calculate and generate a dynamic activity correction coefficient; The benchmark evaluation value is weighted and corrected using the directional weighting coefficient and the dynamic activity correction coefficient to calculate the comprehensive maturation index, which comprehensively characterizes the intensity of the reaction inside the pile. The comprehensive decomposition index is compared with the preset graded decomposition threshold range, and the corresponding pile decomposition identification result is output based on the comparison result.

[0014] Preferably, growth rate analysis is performed based on the detection results to calculate and generate a dynamic activity correction coefficient, including: In response to the detection of the longitudinal penetration enhancement marker, the average link length of the current monitoring period and the historical average link length are retrieved, and the ratio of the difference between the two to the historical average link length is calculated to obtain the longitudinal growth ratio; the longitudinal growth ratio is multiplied by a preset basic longitudinal gain constant to generate a longitudinal active gain value. In response to the detection of the new marker for lateral diffusion, the link quantity characteristics of the current monitoring period and the historical link quantity characteristics are retrieved, and the ratio of the difference between the two to the historical link quantity characteristics is calculated to obtain the lateral diffusion ratio; the lateral diffusion ratio is multiplied by a preset basic lateral gain constant to generate a lateral active gain value. The superposition operation logic is executed to accumulate the longitudinal active gain value and the lateral active gain value to obtain the dynamic active correction coefficient; if none of the above markers are detected, the dynamic active correction coefficient is set to zero or a preset basic balance constant.

[0015] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention constructs a decomposition and infiltration path chain, which imitates the physical process of vegetable decomposition liquid flowing downward along the gaps in the pile under the pull of gravity. It connects discrete dark spots into physical links. By counting the number and length of these links, the system can distinguish between dry spots that are only on the surface and severely rotten channels that have penetrated into the gaps, so that the identification results are more consistent with the real biological reaction state inside the pile. (2) This invention uses timestamp index to associate and store historical data. By calculating the link length growth rate and quantity diffusion rate between the current period and the previous period, it can capture the expansion rate of the decay area, thereby identifying where the new diffusion center is reacting violently. This guides the production line to increase the turning frequency or ventilation in time at the beginning of the decay outbreak to avoid the entire pile of fertilizer from deteriorating. (3) The present invention calculates the direction of vegetable raw material decay extension in the composting area through the diffusion direction analysis module, and gives higher weight to those vertically downward penetration traces, so that the system will not be confused by the shallow decay area with large surface area but less harm, but can identify those vertical decay points that are rooting into the depth of the pile, thus enabling a more scientific assessment of the pile risk. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall system architecture in one embodiment of the present invention.

[0017] In the diagram: 1. Stack image acquisition module; 2. Permeation path construction module; 3. Diffusion direction analysis module; 4. Diffusion feature quantification module; 5. Maturity determination module. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1, please refer to Figure 1 This invention provides a technical solution: an online identification system for the compostability of vegetables used in organic fertilizer production based on machine vision, comprising: The pile image acquisition module 1 is used to acquire surface images of vegetable raw material piles in the organic fertilizer composting area according to a preset time monitoring cycle, and generate timestamp indexes corresponding to the acquisition time; locate rotten areas based on color differences in surface images, and extract the distribution data of mature dark spots associated with the timestamp index. The infiltration path construction module 2 is used to analyze the distribution data of the maturation dark spots, connect the spatially adjacent pixels one by one, simulate the flow traces of the corrosive liquid in the gaps of the pile body, and construct the maturation infiltration path chain. The diffusion direction analysis module 3 is used to analyze the extension trend of each decay and penetration path chain and select the direction with the highest proportion as the dominant decay diffusion direction. The diffusion feature quantification module 4 is used to count the number and average length of the decomposition and infiltration pathway chains, and calculate the product of the two as the total decomposition feature; it generates a longitudinal infiltration enhancement marker based on the increase in the length of the decomposition and infiltration pathway chain, and generates a new transverse diffusion marker based on the increase in the number of decomposition and infiltration pathway chains. The maturity determination module 5 is used to combine the dominant diffusion direction of decay, the total decay characteristics, and the longitudinal penetration enhancement markers and the newly added transverse diffusion markers to evaluate the structured spread state of the decay area on the surface of the pile body and generate the pile body decay recognition result.

[0020] In this invention, the timestamp index generated by the pile image acquisition module 1 serves as the sole time reference for subsequent dynamic trend analysis, ensuring the system can accurately trace back to historical states. The distribution data of decaying dark spots is not a simple collection of image pixels, but rather specific frequency feature data extracted after color space conversion, specifically reflecting the browning substances released after the rupture of vegetable tissue cells. The penetration path construction module 2 uses computer graphics algorithms to reconstruct the liquid flow logic in three-dimensional space on a two-dimensional image, treating discrete spots as nodes through which decaying juice flows. The diffusion direction analysis module 3 aims to identify the macroscopic spread trend of decay, such as determining whether decay is vertically penetrating downwards due to gravity or spreading horizontally due to ventilation equipment. The diffusion feature quantification module 4 transforms abstract image features into specific physical quantities, that is, representing the breadth of decayed area through the number of paths and the depth of decay penetration through the length of paths. The decay maturity determination module 5, by considering the current amount of decay, also combines the growth rate and spread direction of decay to output the pile decay maturity identification result.

[0021] In an optional embodiment, the decayed area is located based on the color difference of the surface image, and the distribution data of decayed dark spots with associated timestamp index is extracted, including: The surface image is converted to the HSV color space, the luminance component and the lightness component are separated, and single-channel grayscale data representing the depth of decay color is extracted. It should be noted that the RGB color space is greatly affected by light intensity. In the actual production environment of a factory, the changes in day and night light can lead to unstable recognition. Therefore, the HSV color space is used to separate color information from brightness information, ensuring that the recognition process only focuses on the color essence of the vegetable surface. Single-channel grayscale data is essentially an enhanced view of rotten features. High grayscale values ​​directly correspond to the dark brown or black features of rotten areas. For example, in the HSV space, even rotten leaves with shadows can still have their color components clearly extracted and will not be misjudged as ordinary shadows. Set a decay grayscale segmentation threshold, perform binarization processing on single-channel grayscale data, and mark pixels whose grayscale values ​​meet the preset decay conditions as dark spot candidate pixels; It should be noted that the grayscale segmentation threshold for decay is a digital boundary that distinguishes normal raw materials from decayed parts. The system compares the grayscale value of each pixel in the image with this threshold and marks the dark pixels that meet the decay characteristics. For example, assuming that the range of single-channel grayscale data is 0 to 255, where the lower the value, the darker the color, if the segmentation threshold calculated by the system is 90, then the pixels in the image corresponding to fresh cabbage leaves with a grayscale value of 180 will be judged as background and removed because they are higher than the threshold. On the other hand, the pixels corresponding to browned and decayed areas with a grayscale value of 45 will be locked by the system and assigned a logical value of "1" or white highlight because they are lower than the threshold and meet the preset decay conditions. This instantly transforms the continuously changing grayscale image into a black and white binary image, thus separating each tiny decay spot from the complex pile background. Morphological opening is performed on the region composed of candidate pixels of dark spots to eliminate discrete isolated noise and smooth the region edges, generating a dark spot binarization mask. It should be noted that morphological opening is an image processing technique that simulates physical filtering. It first removes tiny, isolated spots in the image caused by dust, insects, or camera noise through erosion, and then restores the original outline of the decayed area and smooths the jagged edges through dilation. For example, if there is a black noise spot in the image caused by a mosquito, it will be directly erased after opening, while the real decayed area will be preserved and smoothed out. Traverse the binary mask of dark spots, extract the pixel coordinate index of all mask areas, and associate and store the pixel coordinate index with the timestamp index to generate the distribution data of decayed dark spots; It should be noted that the system extracts the location coordinates of each decayed pixel by traversing the binary mask of the dark spot, and binds and stores these location coordinates with the timestamp of the acquired image; this is equivalent to establishing a spatiotemporal archive of the decayed area. The system not only knows where decay occurs, but also accurately records when it was observed, providing a data foundation for subsequent calculation of the spread speed and path of decay.

[0022] In an optional embodiment, setting a grayscale segmentation threshold for fermentation includes: Statistically analyze the pixel frequency of each gray level in single-channel grayscale data and construct a grayscale frequency histogram. It should be noted that the grayscale frequency histogram is a statistical chart of image brightness. It categorizes and statistically analyzes all pixels in an image according to their brightness levels, intuitively displaying the area ratio of different brightness regions in the image. For example, in a pile image with a resolution of 1000x1000, the system will count that there are 50,000 dark pixels with a grayscale value of 30 and 800,000 light pixels with a grayscale value of 200. Through this statistical method, the originally chaotic image data is transformed into a curve that reflects the distribution pattern of light and dark areas in the pile. Gaussian smoothing is applied to the grayscale frequency histogram to eliminate local noise fluctuations, and the smoothed peak feature points are extracted. It should be noted that Gaussian smoothing uses a Gaussian kernel function to perform convolution operations on the histogram curve. Its purpose is to filter out histogram spikes caused by sensor thermal noise or unevenness in the fine texture of the vegetable surface, so that the statistical curve presents a smooth peak and trough shape, preventing the algorithm from misjudging the peak position due to small fluctuations. For example, the original histogram curve may be full of tiny spikes like a sawtooth, and the system has difficulty distinguishing which is the real peak. After Gaussian smoothing, these sawtooth interferences are filtered out, making the small peaks representing rotten areas and the large peaks representing normal areas clearly visible. Identify the background wave peaks and the decay target wave peaks among the wave peak feature points, and locate the minimum point of the wave trough located between the background wave peaks and the decay target wave peaks; It should be noted that typical vegetable pile images usually exhibit a significant bimodal distribution. The peaks with a large number of pixels and high grayscale represent large areas of fresh vegetable background, while the peaks with a relatively small number of pixels and low grayscale represent rotten targets. The minimum point of the trough between the two is the statistically optimal segmentation point. For example, on the histogram, the tall peaks around grayscale value 200 correspond to large areas of fresh cabbage leaves, while the low peaks around grayscale value 50 correspond to locally browned and rotten stems and leaves. The system will automatically find the deepest depression between these two peaks, namely the trough at grayscale value 120, and lock it as the optimal segmentation point. Read the gray value corresponding to the minimum point of the trough and assign it as the gray value segmentation threshold for subsequent processing; It should be noted that using the grayscale value corresponding to the trough as the segmentation threshold can adaptively handle situations where different batches of vegetable raw materials have varying colors or slight changes in lighting conditions. This ensures that the system always selects the critical value that best balances the foreground and background, avoiding oversegmentation or undersegmentation problems caused by using fixed empirical thresholds. For example, when monitoring a pile of darker-colored eggplant stalks, the overall histogram is biased to the left, and the trough may be located at a grayscale value of 80. However, when monitoring a pile of lighter-colored cabbage, the overall histogram is biased to the right, and the trough may be located at a grayscale value of 150. By reading the trough value, the system can automatically adapt to this difference and accurately separate the rotten parts of each.

[0023] In an optional embodiment, the distribution data of the maturation dark spots is analyzed, and spatially adjacent pixels are connected step by step to simulate the flow traces of the corrosive liquid in the gaps of the pile body, thus constructing a maturation and infiltration path chain, including: Extract the two-dimensional position coordinates of all dark spot pixels in the distribution data of decayed dark spots, and establish a set of pixel coordinates to be traversed; It should be noted that all pixels in the image identified as rotten are placed into a set to be processed, which prepares the data foundation for the subsequent path tracing algorithm. For example, if there are 1,000 scattered rotten pixels in the image, the system will extract the (x,y) coordinates of all 1,000 points and store them in a list or set as the input source for the subsequent algorithm. Select any point from the set of pixel coordinates to be traversed as the current traversal node, and retrieve all discrete points within the preset neighborhood radius of the current traversal node to form a neighborhood candidate pixel group; It should be noted that the neighborhood radius is a preset search range parameter that simulates the maximum continuous distance that putrefactive fluid can penetrate in a physical medium. The system draws a circle with the current point as the center and searches for other putrefactive points within the circle. For example, setting the neighborhood radius to 5 pixels means that only putrefactive points that are less than 5 pixels away from the current point are considered to be possibly belonging to the same penetration path as the current point and will be included in the candidate group. Calculate the spatial Euclidean distance between the current traversed node and each pixel in the neighboring candidate pixel group, and lock the pixel with the smallest Euclidean distance as the secondary extension node. It should be noted that Euclidean distance is the straight-line distance between two points in plane geometry. Calculating this distance aims to find the next decay point that is closest in physical location, simulating the physical property that liquids always flow along the shortest path of least resistance. For example, if there are three points A, B, and C in the candidate group, which are 2, 3, and 4 pixels away from the current point respectively, the system will determine that point A, which is 2 pixels away, is the next inevitable point for the liquid flow and lock it as an extension node. Establish a directional connection vector from the current traversed node to the secondary extended node, update the secondary extended node to the new current traversed node, and remove the connected pixels from the set of pixel coordinates to be traversed. It should be noted that the directional connection vector contains not only length information but also direction information. It connects discrete points into directed line segments. At the same time, in order to prevent the path from being calculated repeatedly, the connected points will be removed from the set to be traversed. For example, the system draws an arrow from the current point to point A, records this flow direction, and then uses point A as the new starting point to prepare to find the next target, while removing point A from the "unprocessed list". The above retrieval, calculation and locking steps are repeated until the updated current traversed node has no neighboring candidate pixels within the preset neighborhood radius. The generated series of directional connection vectors connected end to end are combined to obtain the mature penetration path chain. It should be noted that the system iteratively and recursively finds the nearest point and connects it by repeating the above steps until no new decay points can be found in the surrounding area. For example, after dozens of cycles, the system connects the dozens of originally scattered decay points into a winding downward curve. This curve vividly illustrates how the corrosive liquid seeps and spreads from top to bottom along the gaps in the pile body.

[0024] In one optional embodiment, the extension trend of each decay and infiltration path chain is analyzed, and the direction with the highest proportion is selected as the dominant decay diffusion direction, including: Obtain the starting and ending pixel coordinates of each decay and penetration path chain, and construct a penetration trend vector pointing from the starting pixel coordinates to the ending pixel coordinates; It should be noted that the penetration trend vector is a macroscopic vector connecting the start and end points of the path. It ignores the slight bends and twists caused by the gap structure in the middle of the path and directly reflects the final flow direction of the decay path as a whole. For example, suppose a decay path is as winding as a snake, with a total length of 100 pixels, but its starting coordinates are (10,10) and the ending coordinates are (10,110). The system will construct a vertical downward vector, indicating that although the path sways from side to side in the middle, its fundamental trend is penetration from top to bottom. Calculate the angle between the penetration trend vector and the preset horizontal reference axis to obtain the directional angle characteristics of each path chain; It should be noted that the preset horizontal reference axis is usually selected as the horizontal X-axis of the image coordinate system. The included angle value quantifies the degree of inclination of the decay path relative to the horizontal plane. The vector direction is converted into a specific angle value through geometric operations such as the arctangent function. For example, the included angle value calculated by the system for a decay path that flows vertically downward is 90 degrees or 270 degrees, while the included angle of a horizontally spreading path is close to 0 degrees or 180 degrees. Preset several continuous angle statistical intervals covering 0 to 360 degrees, map the orientation angle features of all path chains to the corresponding angle statistical intervals, and count the distribution frequency of path chains contained in each angle statistical interval. It should be noted that the angle statistics interval divides the 360-degree omnidirectional space into several consecutive sector regions. This approach is similar to histogram statistics in statistics, aiming to group paths with similar directions into one category to eliminate the interference of random angles on individual paths. For example, if the system divides the 0-degree to 360-degree range into 36 intervals, with each interval spanning 10 degrees, if one path has an angle of 88 degrees and another has an angle of 92 degrees, they will both be projected and counted in the vertical interval of approximately 80 to 100 degrees, increasing the counter value for that interval. By numerically comparing the distribution frequencies of all path chains, the dominant angle interval corresponding to the maximum frequency is identified. It should be noted that the dominant angle interval represents the flow direction that most decay paths on the surface of the compost pile follow. This direction is often not randomly generated, but determined by the external physical field or the structure of the compost pile. In the composting scenario, it usually points to the direction of gravity or the downward direction of the moisture gradient. For example, statistical results show that among all 36 intervals, the interval representing the vertical direction contains 150 paths, while the interval representing the horizontal direction contains only 20 paths. By comparing the numerical values, the system locks the interval containing 150 paths as the dominant interval. Extract the center angle value of the dominant angle range and determine it as the dominant direction of decay diffusion; It should be noted that extracting the center angle value as the dominant diffusion direction transforms the range into a definite scalar value, providing a quantitative mathematical basis for subsequent judgment on whether the decay has a dangerous vertical deep penetration trend. For example, if the locked dominant range covers the range of 260 degrees to 280 degrees, the system extracts the median of 270 degrees as the final dominant direction. This value clearly indicates that the decay fluid is mainly affected by gravity and is penetrating vertically downwards into the deep layers, rather than just spreading on the surface.

[0025] In an optional embodiment, the number and average length of the decomposition and infiltration pathway chains are statistically analyzed, and their product is calculated as the total decomposition characteristic, including: The total number of decay and infiltration pathway chains is counted as a characteristic of the number of links representing the dispersion of decay areas; It should be noted that the number of links corresponds to the distribution density of decayed areas on the surface of the pile in a physical sense. The more links there are, the more dispersed the decayed points on the surface are. This feature reflects the breadth of decay. For example, if the system counts fifty independent paths that are not connected to each other on the surface of the pile, it means that the surface of the pile is in the early stage of widespread decay with multiple points of decay, rather than the later stage of single-point infiltration. Based on each decomposition and infiltration path chain, extract all the directional connection vectors contained therein, and calculate the single-chain physical extension length of each path chain by accumulating the Euclidean distance; It should be noted that the physical extension length of a single chain is obtained by accumulating the lengths of all tiny connection vectors along the path. It accurately depicts the actual flow distance of the corrosive liquid as it meanders through the gaps in the pile, rather than a simple straight-line distance from the start to the end. For example, the corrosive liquid may form a winding trajectory by bypassing hard straw or stones during the infiltration process. The cumulative calculation can accurately restore the length of these winding parts, thereby accurately assessing the actual flow of the corrosive liquid. Based on the link quantity characteristic, the arithmetic mean of the physical extension length of all single links is performed to obtain the average link length characterizing the decay depth. It should be noted that the arithmetic mean calculation eliminates the statistical interference of individual abnormally long or short paths. The resulting average link length reflects the general penetration depth of the current pile and represents the overall trend of decay developing inward. For example, although a certain path may flow very long due to the smooth flow of the gap, most paths only stay in the shallow layer. The average value calculated by the system will be too small, and it can be accurately determined that the pile as a whole is still in the shallow decay stage. The link quantity feature is associated with the average link length and the timestamp index is stored. Then, the multiplication operation logic is executed to multiply the link quantity feature with the average link length to obtain the total amount of decayed link. It should be noted that in this embodiment, the total decay characteristic refers to the product of the number of decay penetration path chains and their average length, which is used to quantify the total scale of decay in the pile at present. The multiplication operation logic reflects the mathematical idea of ​​volume estimation, that is, the bottom area multiplied by the height equals the volume. Here, the number of links represents the horizontal bottom area of ​​decay, and the average length represents the vertical height of decay. The product of the two, the total decay characteristic, can comprehensively quantify the overall intensity of the decay reaction inside the pile at the current moment and the amount of biomass consumed. For example, the intensity values ​​calculated by ten short paths and one extremely long path may be different, but the intensity values ​​calculated by five medium-length paths and ten short paths may be similar. This is consistent with the fact that the total amount of vegetable raw materials consumed by the two is similar in terms of physical properties, thus providing a unified quantitative standard for process control.

[0026] In an optional embodiment, generating longitudinal permeability enhancement markers based on the increase in the length of the decay permeation path chain includes: Obtain the average link length calculated for the current monitoring period, and obtain the historical average link length calculated for the previous monitoring period; It should be noted that the monitoring cycle is the minimum time span for the system to perform data sampling. This parameter determines the system's sensitivity in capturing the decay process and ensures continuous observation of the fermentation process. The monitoring cycle is adjusted based on the following: monitoring cycle = surface temperature of the pile × 0.5 minutes, with a minimum of 10 minutes when the temperature is ≥50℃ and a maximum of 60 minutes when the temperature is ≤20℃. The historical average link length is the previous moment's state data stored in non-volatile memory. It provides the necessary time dimension reference for current judgment and forms the cornerstone of dynamic analysis. For example, in the high-temperature and rapid fermentation environment of summer, the system sets the monitoring cycle to 15 minutes. Before each new round of calculation begins, the algorithm program will access the time series database and accurately retrieve the average length record written 15 minutes ago as the benchmark value. The average link length of the current monitoring period is compared with the historical average link length. When the current value is greater than the historical value, a vertical penetration enhancement marker is generated. It should be noted that numerical comparison is the core logic for judging the active state of decay. When the current length value exceeds the historical value, it clearly indicates physically that the decaying fluid is being pulled by gravity to break through the original penetration front and further erode into the deep voids of the pile. The longitudinal penetration enhancement marker is a logical state bit used to trigger subsequent control strategies. Once the marker is activated, it indicates that the longitudinal decay inside the pile is accelerating rather than stabilizing. For example, if the average penetration depth at the previous moment was 10 cm after conversion, while the current detection result shows that the depth has reached 12 cm, this significant increase will be captured by the system and a marker will be generated immediately to notify the central control system to increase the aeration intensity at the bottom to prevent the deep material from smelling bad due to lack of oxygen.

[0027] In an optional embodiment, generating new lateral diffusion markers based on the increase in the number of decay-permeation pathway chains includes: Obtain the link quantity characteristics obtained from the current monitoring period statistics, and obtain the historical link quantity characteristics obtained from the previous monitoring period statistics; The link quantity characteristics of the current monitoring period are compared with the historical link quantity characteristics. When the current value is greater than the historical value, a new horizontal diffusion marker is generated. It should be noted that the number of links directly reflects the distribution density and dispersion of rotten areas on the surface of the pile. Time-series comparison of this feature aims to identify whether rot has spread laterally on the surface. When the number of links counted this time is greater than the historical number, it physically indicates that a new, previously non-existent rot initiation point has emerged on the surface of the pile, which means that the unevenness of the fermentation reaction is increasing. The newly added markers for lateral spread are not only early warning signals, but also the basis for guiding fixed-point turning operations, ensuring that operators can intervene in those newly emerging diseased areas in a timely manner. For example, if monitoring data shows that there were 5 independent rotten patches on the surface of the pile in the previous cycle, while 8 patches were identified in this cycle, the 3 newly added patches indicate that the pathogens are spreading to the surrounding healthy vegetable raw materials. Based on this, the system generates a marker, indicating that the area needs to be mechanically turned immediately to stop the spread.

[0028] In an optional embodiment, by combining the dominant decay diffusion direction, total decay characteristics, and longitudinal penetration enhancement markers and lateral diffusion addition markers, the structured spread state of the decay region on the pile surface is evaluated, generating pile maturity identification results, including: A multidimensional mathematical model for assessing decomposition maturity was constructed, using the total decomposition volume as a benchmark assessment value to characterize the current amount of decay. It should be noted that the multidimensional maturity assessment mathematical model is a comprehensive calculation framework that integrates static stock characteristics and dynamic trend characteristics. Its core foundation lies in the benchmark assessment values. The calculation follows the formula. ;In the formula, The representative link quantity characteristic is the total number of decay and infiltration path chains identified in the current monitoring period, which physically corresponds to the lateral distribution breadth of the decay area. This represents the average link length, which is the arithmetic mean of the physical extension lengths of all path links, and physically corresponds to the longitudinal penetration depth of the decayed region. This formula simulates the volume concept of the decayed region through multiplication. For example, if the system detects that there is decayed region on the current pile surface... There are 10 decay paths, and the average penetration depth of these paths is... If the value is 50 mm, then the calculated baseline evaluation value is... The value is 500, which intuitively quantifies the total amount of absolute biomass degradation that has occurred inside the pile at the current moment; Calculate the cosine of the angle between the dominant diffusion direction of decay and the preset vertical gravity axis, and generate a direction weighting coefficient that reflects the vertical penetration ability of the decay liquid under the action of gravity. It should be noted that in this embodiment, the Y-axis of the image coordinate system is set downwards as the positive direction, and the angle of the vertical gravity axis is fixed at 270 degrees. This step uses trigonometric functions to construct the direction weighting coefficients. The calculation logic follows the formula ;in, The angle value representing the dominant direction of decay diffusion extracted from the path chain. This represents the preset vertical axis angle of gravity, typically set to 90 degrees or 270 degrees. The physical meaning of this formula lies in assessing the degree of alignment between the decay path and the direction of gravity, utilizing the monotonicity of the cosine function to assign a higher weight to vertical penetration; for example, when the dominant direction is identified... It is 270 degrees, relative to the gravity axis. When they completely overlap, the angle difference is 0, and the calculated... A value of 1 indicates that the corrosive fluid is penetrating vertically deep, and the system assigns it full weight; while when the dominant direction is horizontal, the angle difference is 90 degrees. A value close to 0 indicates that the decay is only floating on the surface and has little effect on deep fermentation. The presence status of longitudinally penetrating enhanced markers and laterally diffused newly added markers is detected, and growth rate analysis is performed based on the detection results to calculate and generate dynamic activity correction coefficients; It should be noted that the dynamic activity correction coefficient The concept of rate of change over time is introduced, and its calculation formula integrates the fundamental constant and dynamic gain, i.e. ;In the formula, It is the basic equilibrium constant, usually taken as 1, representing the default coefficient when the system is in steady state; and These are logic status bits, corresponding to the vertical penetration enhancement flag and the lateral diffusion addition flag, respectively, used to control the activation and deactivation of the gain term; and The preset empirical gain constant, , The value range is 0.3-0.8, with 0.6-0.8 for leafy vegetables and 0.3-0.5 for stems, used to adjust the weight of the influence of longitudinal and lateral variations on the results; and These are the vertical growth rate and the horizontal diffusion rate, respectively, calculated from the difference between current and historical values; for example, if the system detects an increase in depth, it sets... The calculated longitudinal growth rate is 1. The gain constant is 0.2 (i.e., a 20% increase). When set to 1, the correction factor is... It will increase the base value of 1 by 0.2, making it 1.2, thereby amplifying the final risk assessment level to warn that decay is accelerating. By using directional weighting coefficients and dynamic activity correction coefficients to perform weighted correction operations on the baseline assessment values, a comprehensive maturation index that comprehensively characterizes the intensity of the internal reaction of the reactor body is calculated. It should be noted that the directional weighting coefficient has a higher weight than the dynamic correction coefficient because vertical infiltration has a more significant impact on compost quality. The weighted correction calculation is a process of integrating the three dimensions of "quantity," "direction," and "velocity" into a dimensionless scalar. The calculation formula is as follows: In this model, A base score was provided. As a directional attenuation factor, As a dynamic amplification factor; making the calculated exponent It is no longer a single physical quantity, but a comprehensive score reflecting the intensity of fermentation in the pile, which can objectively evaluate whether the pile has reached the expected process state; for example: assuming a baseline value is calculated based on previous steps. The directional weighting coefficient is 500. The value is 0.9, which is close to vertical, indicating a dynamic and active correction factor. The system performs a multiplication operation with a value of 1.2, ultimately arriving at 540 as the comprehensive index for ripening. This index is higher than the original benchmark value, accurately reflecting the high-risk state of the current reactor body being in vertical penetration and accelerated diffusion; The comprehensive decomposition index is compared with the preset graded decomposition threshold range, and the corresponding pile decomposition identification result is output based on the comparison result. It should be noted that the preset threshold ranges are usually defined based on national organic fertilizer fermentation standards or empirical data from historical best batches, representing the physical characteristic boundaries of different fermentation stages. For example, the system presets values ​​of 0 to 40 for the raw material range during the heating period, 40 to 80 for the decomposition range during the high-temperature period, and 80 to 100 for the fully decomposed range during the aging period. If the calculated comprehensive decomposition index at the current moment is 92, and this value falls precisely within the range of 80 to 100, the system will immediately output the identification result of fully decomposed, and link the control system to light up a green light or automatically send a discharge command, clearly informing the operator that this batch of vegetable raw materials has been successfully converted into qualified organic fertilizer and does not need to continue fermentation.

[0029] In an optional embodiment, growth rate analysis is performed based on the detection results to calculate and generate a dynamic activity correction coefficient, including: In response to the detection of a longitudinal penetration enhancement marker, the average link length of the current monitoring period and the historical average link length are retrieved, and the ratio of the difference between the two to the historical average link length is calculated to obtain the longitudinal growth ratio. The longitudinal growth ratio is then multiplied by a preset basic longitudinal gain constant to generate a longitudinal active gain value. It should be noted that the longitudinal growth ratio is a normalized measure of the rate of change in decay depth. It eliminates the influence of the magnitude of the baseline depth value and focuses on the drasticness of the change. The larger the ratio, the faster the decay penetrates inward. The preset baseline longitudinal gain constant is an empirical weighting coefficient used to adjust the influence of depth changes on the overall activity assessment, preventing fluctuations in a single dimension from dominating the entire assessment result. For example, assuming the average decay depth measured in the previous monitoring period was 10 cm, and the measured depth has rapidly increased to 12 cm this time, the system calculates the difference to be 2 cm. The ratio of this difference to the historical baseline of 10 cm is 0.2. If the preset gain constant is 0.5, the longitudinal activity gain value obtained by multiplying the two is 0.1, which quantitatively indicates that the decay activity of the pile in the vertical direction has increased by 10% at the current moment. In response to the detection of a new marker for lateral diffusion, the link quantity characteristics of the current monitoring period and the historical link quantity characteristics are retrieved, and the ratio of the difference between the two to the historical link quantity characteristics is calculated to obtain the lateral diffusion ratio. The lateral diffusion ratio is then multiplied by a preset basic lateral gain constant to generate the lateral active gain value. It should be noted that the lateral diffusion ratio reflects the explosive force of the expansion of rotten areas on the surface of the pile. By comparing the number of new and old links, the system can capture the proliferation trend of tiny lesions that are difficult to detect with the naked eye. This ratio essentially quantifies the invasiveness of the fermentation community on the horizontal plane. The basic lateral gain constant is an adjustment factor set according to the perishability of different vegetable raw materials, used to balance the proportion of lateral diffusion in the total activity. For example, if the number of independent rotten patches on the surface of the pile increases from 5 at the previous moment to 8 at this moment, the difference is 3. Compared with the original base of 5, the lateral diffusion ratio is as high as 0.6. If the gain constant is set to 0.3 for perishable tomato raw materials, the calculated lateral activity gain value is 0.18. This value warns operators that the surface rot is growing explosively at a relative rate of 60%. The superposition operation logic is executed to accumulate the vertical active gain value and the horizontal active gain value to obtain the dynamic active correction coefficient; if none of the above flags are detected, the dynamic active correction coefficient is set to zero or the preset basic balance constant. It should be noted that the superposition operation logic follows the principle of risk accumulation. Whether the depth or breadth of the superposition increases, it will lead to an increase in the final correction coefficient, thereby improving the rating of the current decay state. This mechanism ensures that the system will not miss any signs of deterioration in any dimension. The operation of setting it to zero or to the basic equilibrium constant plays a filtering and stabilizing role, ensuring that when the fermentation process is stable and there are no drastic changes, the evaluation result output by the system mainly depends on the current static inventory, avoiding false fluctuations in the data. For example, adding the previously calculated vertical gain value of 0.1 and the horizontal gain value of 0.18, we get a dynamic active correction coefficient of 0.28. This coefficient will be used to improve the maturity score, which means that the fermentation reaction is in a period of rapid acceleration. Conversely, if the depth and quantity of decay do not increase during the monitoring period, the system will lock this coefficient directly to 0. At this time, the evaluation is entirely based on the existing decay area and depth, and no dynamic gain is superimposed.

[0030] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A machine vision-based online identification system for the composting degree of vegetables used in organic fertilizer production, characterized in that, include: The pile image acquisition module is used to acquire surface images of vegetable raw material piles in the organic fertilizer composting area according to a preset time monitoring cycle, and generate timestamp indexes corresponding to the acquisition time; locate rotten areas based on the color differences of the surface images, and extract the distribution data of mature dark spots associated with the timestamp index; The infiltration path construction module is used to analyze the distribution data of the decomposed dark spots, connect the spatially adjacent pixels one by one, simulate the flow traces of the decomposed liquid in the gaps of the pile body, and construct the decomposed infiltration path chain. The diffusion direction analysis module is used to analyze the extension trend of each of the aforementioned decay and penetration path chains, and select the direction with the highest proportion as the dominant decay diffusion direction. The diffusion feature quantification module is used to count the number and average length of the decomposition and infiltration path chains, and calculate the product of the two as the total decomposition feature; it generates a longitudinal infiltration enhancement marker based on the increase in the length of the decomposition and infiltration path chain, and generates a lateral diffusion addition marker based on the increase in the number of the decomposition and infiltration path chains. The decay maturity determination module is used to combine the dominant decay diffusion direction, the total decay characteristics, the longitudinal penetration enhancement marker, and the lateral diffusion addition marker to evaluate the structured spread state of the decay area on the surface of the pile body and generate the pile body decay maturity identification result.

2. The online identification system for vegetable composting degree based on machine vision for organic fertilizer production according to claim 1, characterized in that, Based on the color differences in the surface image, the decayed areas are located, and the distribution data of decayed dark spots associated with the timestamp index is extracted, including: The surface image is converted into the HSV color space, the luminance component and the brightness component are separated, and single-channel grayscale data representing the depth of decay color is extracted. Set a decay grayscale segmentation threshold, perform binarization processing on the single-channel grayscale data, and mark pixels whose grayscale values ​​meet the preset decay conditions as dark spot candidate pixels. A morphological opening operation is performed on the region composed of the dark spot candidate pixels to eliminate discrete isolated noise and smooth the region edges, thereby generating a dark spot binarization mask. Traverse the binary mask of dark spots, extract the pixel coordinate indices of all mask regions, and associate and store the pixel coordinate indices with the timestamp index to generate the decayed dark spot distribution data.

3. The online identification system for vegetable composting degree based on machine vision for organic fertilizer production according to claim 2, characterized in that, Set the grayscale segmentation threshold for ripening, including: The frequency of pixel occurrence at each gray level in the single-channel grayscale data is statistically analyzed, and a grayscale frequency histogram is constructed. Gaussian smoothing is applied to the grayscale frequency histogram to eliminate local noise fluctuations and extract the smoothed peak feature points; Identify the background wave peaks and the decay target wave peaks among the wave peak feature points, and locate the minimum point of the valley between the background wave peaks and the decay target wave peaks; Read the gray value corresponding to the minimum point of the trough and assign it as the value of the mature gray value segmentation threshold for subsequent processing.

4. The online identification system for vegetable decomposition degree in organic fertilizer production based on machine vision according to claim 3, characterized in that, Analyzing the distribution data of the decomposition dark spots, connecting spatially adjacent pixels step by step, simulating the flow traces of the decomposition liquid in the gaps of the pile body, and constructing a decomposition and infiltration path chain, including: Extract the two-dimensional position coordinates of all dark spot pixels in the decayed dark spot distribution data, and establish a set of pixel coordinates to be traversed; Select any point from the set of pixel coordinates to be traversed as the current traversal node, and retrieve all discrete points within the preset neighborhood radius of the current traversal node to form a neighborhood candidate pixel group. Calculate the spatial Euclidean distance between the current traversed node and each pixel in the neighboring candidate pixel group, and lock the pixel with the smallest Euclidean distance as the secondary extension node; Establish a directional connection vector from the current traversed node to the secondary extended node, update the secondary extended node to the new current traversed node, and remove the connected pixels from the set of pixel coordinates to be traversed. The above retrieval, calculation, and locking steps are repeated until the updated current traversed node has no neighboring candidate pixels within the preset neighborhood radius. The generated series of directional connection vectors connected end to end are combined to obtain the decay and penetration path chain.

5. The online identification system for vegetable composting degree based on machine vision for organic fertilizer production according to claim 4, characterized in that, Analyze the extension trend of each of the aforementioned decomposition and infiltration pathway chains, and select the trend with the highest proportion as the dominant direction of decay diffusion, including: Obtain the starting and ending pixel coordinates of each of the aforementioned decay and penetration path chains, and construct a penetration trend vector pointing from the starting pixel coordinates to the ending pixel coordinates; Calculate the angle between the penetration trend vector and the preset horizontal reference axis to obtain the directional angle characteristics of each path chain; Several consecutive angular statistical intervals covering 0 to 360 degrees are preset, and the directional angular features of all the path chains are mapped to the corresponding angular statistical intervals. The distribution frequency of the path chains contained in each angular statistical interval is counted. By numerically comparing the distribution frequencies of all the aforementioned path chains, the dominant angle interval corresponding to the maximum frequency value is identified. Extract the center angle value of the dominant angle range and determine it as the dominant diffusion direction of decay.

6. The online identification system for vegetable composting degree based on machine vision for organic fertilizer production according to claim 5, characterized in that, The number and average length of the aforementioned decomposition and infiltration pathway chains are statistically analyzed, and their product is calculated as the total decomposition characteristic, including: The total number of the aforementioned decay and infiltration pathway chains is counted and used as a characteristic of the number of links to characterize the dispersion of decay areas; Based on each of the aforementioned decomposition and infiltration path chains, extract all the directional connection vectors contained therein, and calculate the single-chain physical extension length of each path chain by accumulating Euclidean distances; Based on the link quantity characteristic, the arithmetic mean of the physical extension length of all the single links is calculated to obtain the average link length characterizing the decay depth. The link quantity feature is associated with the average link length and stored using the timestamp index. Then, a multiplication operation is performed to multiply the link quantity feature by the average link length to obtain the total amount of decayed link feature.

7. The online identification system for vegetable composting degree in organic fertilizer production based on machine vision according to claim 6, characterized in that, Based on the increase in the length of the decay and permeation pathway chain, a longitudinal permeation enhancement marker is generated, including: Obtain the average link length calculated for the current monitoring period, and obtain the historical average link length calculated for the previous monitoring period; The average link length of the current monitoring period is compared with the historical average link length. When the current value is greater than the historical value, a vertical penetration enhancement marker is generated.

8. The online identification system for vegetable composting degree based on machine vision for organic fertilizer production according to claim 7, characterized in that, Based on the increase in the number of the decomposition and infiltration pathway chains, new markers for lateral diffusion are generated, including: Obtain the link quantity characteristics obtained from the statistics of the current monitoring period, and obtain the historical link quantity characteristics obtained from the statistics of the previous monitoring period; The link quantity characteristics of the current monitoring period are compared with the historical link quantity characteristics. When the current value is greater than the historical value, a new horizontal diffusion marker is generated.

9. The online identification system for vegetable decomposition degree in organic fertilizer production based on machine vision according to claim 8, characterized in that, Combining the dominant decay diffusion direction, the total decay characteristics, the longitudinal penetration enhancement markers, and the lateral diffusion addition markers, the structured spread state of the decay region on the pile surface is evaluated, generating pile maturity identification results, including: A multidimensional mathematical model for assessing decomposition is constructed, and the total decomposition characteristics are used as the benchmark assessment values ​​to characterize the current amount of decay. Calculate the cosine of the angle between the dominant decay diffusion direction and the preset vertical gravity axis to generate a direction weighting coefficient that reflects the vertical penetration ability of the decay liquid under the action of gravity. The presence status of the longitudinal penetration enhancement marker and the lateral diffusion addition marker is detected, and growth rate analysis is performed based on the detection results to calculate and generate a dynamic activity correction coefficient; The benchmark evaluation value is weighted and corrected using the directional weighting coefficient and the dynamic activity correction coefficient to calculate the comprehensive maturation index, which comprehensively characterizes the intensity of the reaction inside the pile. The comprehensive decomposition index is compared with the preset graded decomposition threshold range, and the corresponding pile decomposition identification result is output based on the comparison result.

10. The online identification system for vegetable decomposition degree in organic fertilizer production based on machine vision according to claim 9, characterized in that, Based on the detection results, growth rate analysis is performed to calculate and generate dynamic activity correction coefficients, including: In response to the detection of the longitudinal penetration enhancement marker, the average link length of the current monitoring period and the historical average link length are retrieved, and the ratio of the difference between the two to the historical average link length is calculated to obtain the longitudinal growth ratio; the longitudinal growth ratio is multiplied by a preset basic longitudinal gain constant to generate a longitudinal active gain value. In response to the detection of the new marker for lateral diffusion, the link quantity characteristics of the current monitoring period and the historical link quantity characteristics are retrieved, and the ratio of the difference between the two to the historical link quantity characteristics is calculated to obtain the lateral diffusion ratio; the lateral diffusion ratio is multiplied by a preset basic lateral gain constant to generate a lateral active gain value. The superposition operation logic is executed to accumulate the longitudinal active gain value and the lateral active gain value to obtain the dynamic active correction coefficient; if none of the above markers are detected, the dynamic active correction coefficient is set to zero or a preset basic balance constant.

Images