A method for monitoring a fruit bottom colored reflective film

Abstract

The present application belongs to the technical field of intelligent agriculture, and discloses a monitoring method for a fruit bottom colorizing and light reflecting film, comprising the following steps: S1, forming a grid map for an orchard; S2, collecting fruit bottom images to obtain array type multi-angle image data; S3, performing edge processing on the image data to form fruit images; S4, performing fruit surface color analysis on the fruit images to form color feature data; S5, performing edge and dispersion analysis on dark color patches in the color analysis process to form lesion feature data; S6, inputting the color feature data and the lesion feature data of S4 and S5 into a preset fruit maturity and lesion grade library to obtain preliminary maturity and sunscald grades; S7, performing feature weighting judgment to output maturity grades and / or sunscald lesion probabilities; S8, if the lesion probability is greater than a preset threshold, automatically recording a collection time stamp, a lesion probability value, an image slice and a map location, and triggering an early warning of adjusting the light reflecting film to weaken the radiant light; if the lesion probability is less than or equal to the preset threshold, normally recording data.

Description

Technical Field

[0001] This invention belongs to the field of intelligent agricultural technology, specifically relating to a monitoring method for a coloring reflective film on the bottom of fruit. Background Technology

[0002] Reflective film is primarily used in orchards for cash crops such as apples, grapes, citrus, and peaches, where high-quality fruit color is crucial. Its application has become a key step in producing high-end commercial fruit. By reflecting sunlight of a specific spectrum, reflective film improves light conditions in the lower and inner parts of the tree canopy, promoting even fruit coloring and, to some extent, increasing sugar content and firmness. This directly enhances the commercial grade of the fruit and the overall profitability of the orchard.

[0003] Due to differences in the location and growth of fruit trees, individual fruits exhibit variations. In some areas, the sunlight emitted by the reflective film is too strong, causing sunburn on the fruit. The reflective film needs to be adjusted promptly to weaken the emitted light. In other areas, the bottom color of the fruit is not up to standard, requiring an extension of the application time. The current method involves personnel periodically inspecting the fruit on-site. This is not only inefficient but also risks trampling or even damaging the reflective film, affecting its effectiveness. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a monitoring method for reflective films used to color the bottom of fruits, eliminating the need for regular on-site inspections. The method is as follows: A method for monitoring reflective films used for bottom coloring of fruit includes the following steps: S1. Locate the fruit trees in the orchard, create a grid map, and store it in the database; S2. Under the active dimming light field composed of a camera and LED lights, the machine runs and collects images of the bottom of the fruit to obtain array-type multi-angle image data. S3. Edge processing of image data to accurately segment the fruit target and form a fruit image against a complex background of branches and leaves; S4. Perform color analysis on the surface of the fruit in the fruit image to generate color feature data; S5. During the color analysis process, the edges and dispersion of dark patches are analyzed to form lesion characteristic data; S6. Input the color feature data and lesion feature data from S4 and S5 into the preset fruit maturity and lesion grade database to obtain the preliminary maturity and sunburn grade. S7. Perform feature-weighted judgment on the preliminary maturity level and sunburn level, and output the maturity level and the probability of sunburn lesions. S8. If the probability of sunburn lesions is greater than the preset threshold, the collection timestamp, lesion probability value, image slice, and location on the map will be automatically recorded, triggering an early warning that the reflective film needs to be adjusted to weaken the radiation. At the same time, the early warning signal can be directly used as a control command to trigger the corresponding area to perform actions. If the probability of lesions is less than or equal to the preset threshold, the data will be recorded normally.

[0005] Furthermore, the edge-processing in step S3 includes the following steps: S31. Obtain the variety information of the fruit trees, and select the corresponding color space components according to the variety information; S32. Based on the color space components, an edge detection algorithm is used to extract the edge information between the fruit and the background; S33. Perform shape fitting on the edge information to identify and locate candidate fruit regions with circular or near-circular outline features; S34. Using the candidate fruit region as a constraint, the image is segmented to extract a fruit mask image containing only the fruit target, thus forming the fruit image.

[0006] Furthermore, the color analysis in step S4 specifically includes the following steps: S41. Convert the fruit image to a hue-saturation-brightness color space; S42. Divide the fruit image into multiple sub-regions; S43. Calculate the hue channel values, saturation channel values, and area proportion of specific hue pixels in each sub-region that reflect fruit maturity. S44. Calculate the overall coloring uniformity of the fruit based on the area ratio of each sub-region and the preset region weight. S45. Combine the hue channel value, saturation channel value, and color uniformity to form the color feature data.

[0007] Furthermore, step S5, the analysis of the edges and dispersion of dark patches, includes the following steps: S51. Extract the micro-texture feature vector representing the local texture change pattern within the fruit image area; S52. Extract the edge sharpness feature vector that characterizes the degree of drastic change in the edge gradient of the patch; S53. Identify suspected lesion areas on the fruit surface that differ from the main color based on a color clustering algorithm; S54. Calculate the spatial dispersion index of the color distribution within the suspected lesion area; S55. The micro-texture feature vector, the edge sharpness feature vector, and the spatial dispersion index are combined to form the lesion feature data.

[0008] Furthermore, the fruit maturity and lesion grade database includes fruit maturity grades and lesion grades for different varieties.

[0009] Furthermore, the feature weighting formula includes a maturity weighting formula and a sunburn lesion probability weighting formula; The maturity weighting formula is as follows: in, Output value for maturity level. This is a preliminary maturity quantification score based on the color feature data mapping. This is the quantification score for the color uniformity. and Preset weighting coefficients; The weighted formula for the probability of sunburn lesions is as follows: in, The probability of sunburn lesions. The mean of the irregularity of the edges of all dark patches. The spatial dispersion index is mentioned above. , , These are the model parameters determined through sample training.

[0010] Compared with the prior art, the advantages of the present invention are as follows: Compared with existing technologies, this invention, through its monitoring design of a colored reflective film on the bottom of the fruit, compensates for the blind spots of traditional detection methods. It can promptly capture the coloring of the bottom of the fruit, accurately identify areas of abnormal strong light, and achieve early warning of sunburn risk and location of lesions. Combined with model parameters trained on samples, it further improves the accuracy and efficiency of detection, provides reliable support for post-harvest quality grading and control of fruit, effectively reduces the risk of defective fruit flowing out, and improves the quality control level of the fruit industry. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating the monitoring method of an embodiment of the present invention. Detailed Implementation

[0012] To facilitate understanding of the present invention by those skilled in the art, specific embodiments of the present invention will be described below with reference to the accompanying drawings.

[0013] like Figure 1 As shown, this invention proposes a monitoring method for a coloring reflective film on the bottom of fruit, comprising the following steps: Step S1: Locate the fruit trees in the orchard, create a grid map, and store it in the database.

[0014] The primary application scenario for this application is when reflective film has already been laid in the orchard. To improve the reflective effect of the film, frequent personnel movement should be minimized. Therefore, the fruit trees in the orchard are located in advance to form a grid map. Once an anomaly is detected during monitoring, the location coordinates of the anomaly point can be sent to the system, allowing staff to promptly obtain the accurate location and reduce frequent movement.

[0015] Step S2: Under the active dimming light field composed of the camera and LED lights, the machine runs and collects images of the bottom of the fruit to obtain array-type multi-angle image data.

[0016] Specifically, as the machine moves forward in a straight line, two cameras on the machine take turns continuously capturing images from above, obtaining an array of images of the fruit trees above. During the image acquisition process, the light sensor adjusts the brightness of the LED lights according to the changes in the light under the trees to counteract the interference of the complex light environment on image acquisition.

[0017] Step S3: Edge processing of image data to accurately segment the fruit target and form a fruit image against a complex background of branches and leaves.

[0018] Edge-out processing includes the following steps: S31. Obtain the variety information of the fruit trees and select the corresponding color space components based on the variety information; Specifically, information about fruit tree varieties can be pre-stored or obtained through identification. For example, for red varieties, the a channel of the Lab color space can be preferentially used to enhance the contrast between the fruit and the green leaves; for yellow-green varieties, the H and V channels of the HSV color space can be combined.

[0019] S32. Based on color space components, an edge detection algorithm is used to extract the edge information between the fruit and the background; Specifically, by using existing edge detection algorithms and dynamically adjusting the high and low thresholds of edge detection using the local contrast mean, adaptive edge extraction can be achieved to overcome the impact of uneven lighting under the forest canopy on edge continuity.

[0020] In this embodiment, the edge detection algorithm used is adaptive Canny edge detection. Specifically, the average gradient magnitude within a local window with a side length of 1 / 5 of the estimated fruit diameter is first calculated. Then follow the formula and The low and high thresholds are dynamically determined, where k is an adjustment coefficient between 0.2 and 0.4, adjusted in real time based on feedback from the light sensor. This method effectively overcomes the impact of uneven lighting in the forest canopy on edge continuity, suppressing noise interference while maintaining edge detection sensitivity.

[0021] S33. Perform shape fitting on the edge information to identify and locate candidate fruit regions with circular or near-circular contour features; Specifically, the shape fitting is a circular target fitting, which uses a shape detection algorithm to select connected regions that conform to the fruit contour features as foreground labels.

[0022] In this embodiment, the shape detection algorithm used is the improved probabilistic Hough circle detection. Specifically, the edge image is first preprocessed to remove isolated edge segments shorter than a preset threshold and regions with disordered gradient directions; then, the remaining edge points are subjected to Hough transform voting, and a roundness coefficient is introduced for the candidate circles obtained from the voting. (Where A is the area of ​​the connected region and P is the perimeter) is used as the screening criterion, retaining only connected regions with C > 0.7 and edge point continuity of no less than 30% as candidate fruit targets; the radius range is adaptively expanded according to the preset value range of the current variety based on the 3σ principle to cover the size variations of young and extra-large fruits. This method can stably locate fruit regions with circular or near-circular outlines under complex branch and leaf shading conditions.

[0023] S34. Using the candidate fruit region as a constraint, the image is segmented to extract a fruit mask image containing only the fruit target, thus forming the fruit image.

[0024] Specifically, based on foreground markers, a marker-controlled segmentation algorithm is used to process adhered and occluded edges, ultimately generating an accurate fruit mask image.

[0025] In this embodiment, the segmentation algorithm used is marker-controlled watershed segmentation. Specifically, the candidate fruit regions obtained from the improved probabilistic Hough circle detection in step S33 are used as foreground markers, and the regions outside the image boundaries are used as background markers. The gradient magnitude image is corrected through distance transformation to suppress the generation of local minima. Finally, a watershed transformation is performed to complete the fine segmentation. This method can effectively handle the problems of fruit adhesion and edge breakage caused by foliage occlusion, ultimately outputting a precise binary mask image containing only the fruit targets.

[0026] Step S4: Perform color analysis on the surface of the fruit in the fruit image to generate color feature data.

[0027] Color analysis includes the following steps: S41. Convert the fruit image to the hue-saturation-brightness color space; In this embodiment, the fruit image is converted to HSV space, and the H and S channel data are extracted. The purpose of this conversion is to separate color information from brightness information, reduce the interference of illumination changes on color feature extraction, and more stably extract hue and saturation features that reflect fruit ripeness.

[0028] S42. Divide the fruit image into multiple sub-regions; In this embodiment, the fruit area is divided into grid-like sub-regions (such as a nine-square grid).

[0029] S43. Calculate the hue channel values, saturation channel values, and area proportion of specific hue pixels in each sub-region that reflect fruit maturity. Specifically, calculate the average hue, average saturation, and pixel percentage of the target hue (such as red) for each partition.

[0030] S44. Calculate the overall coloring uniformity of the fruit based on the area ratio of each sub-region and the preset region weight. Specifically, the preset regional weights are based on the principle of high weight for central regions and low weight for peripheral regions.

[0031] S45. Combine the hue channel value, saturation channel value, and color uniformity to form the color feature data.

[0032] Step S5: During the color analysis process, the edges and dispersion of dark patches are analyzed to form lesion characteristic data.

[0033] Further analysis of the edges and dispersion of dark patches includes the following steps: S51. Analyze the epidermal texture and extract the micro-texture feature vector representing the local texture change pattern within the fruit image area. S52. Analyze the gradient change characteristics of the dark area edges. Sunburned areas usually have blurred edges and low sharpness. Extract the edge sharpness feature vector that represents the degree of drastic gradient change of the patch edges. S53. Identify suspected lesion areas on the fruit surface that differ from the main color based on a color clustering algorithm; Color clustering algorithms are used to cluster and segment pixels in a color space to identify tiny regions that are significantly different in color from healthy fruit surfaces as candidate lesions. S54. Calculate the spatial dispersion index of color distribution within the suspected lesion area, i.e., assess whether the distribution of candidate lesions on the fruit surface is clustered or dispersed. When dark patches are clustered, the spatial dispersion index value is small, indicating that the lesion area is concentrated, consistent with the spatial characteristics of sunburn lesions usually formed by localized burning from strong light on reflective film. When dark patches are dispersed, the spatial dispersion index value is large, indicating that the suspected patches are scattered, possibly caused by natural spots on the fruit surface or non-sunburn factors. In the lesion grade matching in step S6 and the weighted calculation in step S7, the higher the clustering (the lower the dispersion), the greater the corresponding sunburn lesion risk weight.

[0034] S55. Combine the micro-texture feature vector, edge sharpness feature vector, and spatial dispersion index to form lesion feature data, that is, generate lesion feature data by integrating texture, edge, and dispersion data.

[0035] Step S6: Input the color feature data and lesion feature data from S4 and S5 into the preset fruit maturity and lesion grade database to obtain the preliminary maturity and sunscald grade.

[0036] Specifically, the fruit maturity and lesion grade databases are pre-constructed in the following ways: (1) Construction of maturity database: Collect bottom sample images of fruits at different maturity stages (unripe, color change, suitable for harvesting, fully ripe) according to variety, manually label the maturity level, extract and statistically analyze the color feature data distribution range of each sample, and establish a variety-maturity mapping threshold table.

[0037] (2) Construction of disease spot grade database: Collect fruit sample images of different sunburn severity (no, light, medium and heavy) according to variety, manually label the disease spot grade, extract and statistically analyze the distribution range of disease spot characteristic data (such as edge roughness and dispersion) of each sample, and establish a variety-disease spot grade mapping threshold table.

[0038] (3) After associating the above two tables with the variety identification information, store them in the database for query and comparison in step S6.

[0039] Step S7: Perform feature-weighted judgment on the preliminary maturity level and sunburn level, and output the maturity level and the probability of sunburn lesions.

[0040] (1) Calculation of maturity level output value: The preliminary maturity quantification score obtained by querying the maturity level database in step S6 is denoted as... The color uniformity index calculated in step S4 is denoted as Both are normalized to a numerical range of 0-100. The maturity level output value M is calculated using the following formula:

[0041] In the formula, w1 and w2 are preset weight coefficients, satisfying w1 + w2 = 1. The weight coefficients can be adaptively adjusted according to the differences in coloring characteristics of the varieties. For example, for red varieties, the color characteristic weight w1 can be set to 0.6, and the coloring uniformity weight w2 can be set to 0.4. Finally, based on the preset threshold range where the M value is located, the maturity level (such as immature stage, color-changing stage, suitable harvesting stage, or fully mature stage) is mapped and output.

[0042] (2) Calculation of the probability of sunburn lesions: Two key indicators were extracted from the lesion feature data obtained in step S5: the average degree of irregularity at the edges of all dark patches. And the spatial distribution dispersion index D of dark patches on the fruit surface. The probability S of sunburn lesions is calculated using the following formula: ; In the formula, α and β are positive coefficients, and γ is the offset threshold. The above parameters are predetermined by acquiring standard sample images of different sunscald severity levels and extracting the values ​​from each sample. Using the D value and manually labeled sunscald level as labels, a logistic regression model is used for training and fitting to obtain the optimal combination of α, β, and γ parameters, which are then stored in a database for each variety for later retrieval. Preferably, before training, the parameters of each sample are... The eigenvalues ​​of D are normalized to eliminate the influence of dimensional differences on model convergence; in the actual calculation of the probability of sunburn lesions, the eigenvalues ​​of the fruit to be tested are also normalized. Perform the same normalization process as for D. The output value of S ranges from 0 to 1. Multiplying S by 100% gives the final probability of sunburn lesions.

[0043] Step S8: If the probability of sunburn lesions is greater than the preset threshold, automatically record the collection timestamp, lesion probability value, image slice, and location on the map, triggering an early warning that the reflective film needs to be adjusted to weaken radiation. At the same time, this early warning signal can be directly used as a control command to trigger the corresponding area to perform actions to achieve regulation. If the probability of lesions is less than or equal to the preset threshold, record data normally.

[0044] The control command is an automated data update command directly converted from the warning signal in step S8. The core of this command is to automatically record and update the location information and related data of fruits whose probability of sunburn disease exceeds a preset threshold to the system's database.

[0045] The corresponding area refers to the specific grid location in the gridded map established in step S1, precisely locked by the "location on the map" recorded in step S8, where sunscald lesions have occurred. This area is the coordinate identifier of the specific fruit tree where the problematic fruit is located on the database map.

[0046] The execution action refers to incorporating the lesion information of the aforementioned area into the database. Specifically, the system automatically executes a data recording action, taking the key information of this detection (including the automatically recorded acquisition timestamp, lesion probability value, image slice, and location coordinates on the gridded map) as a new lesion record for that area, and writing it completely into the system's database for storage and association, providing clear data support for statistical analysis.

[0047] The embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for monitoring a reflective film used for coloring the bottom of fruit, characterized in that, Includes the following steps: S1. Locate the fruit trees in the orchard, create a grid map, and store it in the database; S2. Under the active dimming light field composed of a camera and LED lights, the machine runs and collects images of the bottom of the fruit to obtain array-type multi-angle image data. S3. Edge processing of image data to accurately segment the fruit target and form a fruit image against a complex background of branches and leaves; S4. Perform color analysis on the surface of the fruit in the fruit image to generate color feature data; S5. During the color analysis process, the edges and dispersion of dark patches are analyzed to form lesion characteristic data; S6. Input the color feature data and lesion feature data from S4 and S5 into the preset fruit maturity and lesion grade database to obtain the preliminary maturity and sunburn grade. S7. Perform feature-weighted judgment on the preliminary maturity level and sunburn level, and output the maturity level and the probability of sunburn lesions. S8. If the probability of sunburn lesions is greater than the preset threshold, automatically record the collection timestamp, lesion probability value, image slice, and location on the map, triggering an early warning that the reflective film needs to be adjusted to weaken the radiation; if the probability of lesions is less than or equal to the preset threshold, record data normally.

2. The monitoring method according to claim 1, characterized in that, The edge-processing in step S3 includes the following steps: S31. Obtain the variety information of the fruit trees, and select the corresponding color space components according to the variety information; S32. Based on the color space components, an edge detection algorithm is used to extract the edge information between the fruit and the background; S33. Perform shape fitting on the edge information to identify and locate candidate fruit regions with circular or near-circular outline features; S34. Using the candidate fruit region as a constraint, the image is segmented to extract a fruit mask image containing only the fruit target, thus forming the fruit image.

3. The monitoring method according to claim 1, characterized in that, The color analysis in step S4 specifically includes the following steps: S41. Convert the fruit image to a hue-saturation-brightness color space; S42. Divide the fruit image into multiple sub-regions; S43. Calculate the hue channel values, saturation channel values, and area proportion of specific hue pixels in each sub-region that reflect fruit maturity. S44. Calculate the overall coloring uniformity of the fruit based on the area ratio of each sub-region and the preset region weight. S45. Combine the hue channel value, saturation channel value, and color uniformity to form the color feature data.

4. The monitoring method according to claim 1, characterized in that, Step S5, the analysis of the edges and dispersion of dark patches, includes the following steps: S51. Within the fruit image area, extract the micro-texture feature vector that characterizes the local texture change pattern; S52. Extract the edge sharpness feature vector that characterizes the degree of drastic change in the edge gradient of the patch; S53. Identify suspected lesion areas on the fruit surface that differ from the main color based on a color clustering algorithm; S54. Calculate the spatial dispersion index of the color distribution within the suspected lesion area; S55. The micro-texture feature vector, the edge sharpness feature vector, and the spatial dispersion index are combined to form the lesion feature data.

5. The monitoring method according to claim 1, characterized in that, The fruit maturity and lesion grade database includes fruit maturity grades and lesion grades for different varieties.

6. The monitoring method according to claim 1, characterized in that, The feature weighting formulas include the maturity weighting formula and the sunburn lesion probability weighting formula; The maturity weighting formula is as follows: in, Output value for maturity level. This is a preliminary maturity quantification score based on the color feature data mapping. This is the quantification score for the color uniformity. and Preset weighting coefficients; The weighted formula for the probability of sunburn lesions is as follows: in, The probability of sunburn lesions. The mean of the irregularity of the edges of all dark patches. The spatial dispersion index is mentioned above. , , These are the model parameters determined through sample training.

Images