A method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits.
By using image processing and temperature correction techniques, key characteristics of flower and fruit growth status are identified, and yield forecasts are dynamically adjusted. This solves the problem of the disconnect between yield prediction and actual growth status in existing technologies, and achieves accurate quantitative assessment of agricultural product yield.
Patent Information
- Application Number
- CN202611195019.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-07
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies cannot dynamically reflect key dynamic changes in the growth process of flowers and fruits, lack the quantification of visual characteristics of flower and fruit growth status and the coupling of cross-phenological data, resulting in a disconnect between yield prediction and actual growth status, and making it impossible to achieve accurate quantitative yield assessment.
By acquiring images of the full bloom and young fruit stages, color space segmentation and Hough circle transform detection are used to identify the initial flower base, competitive weak fruit, and effective base. The flower base is corrected by combining temperature data, and pixel and texture features of young fruit are extracted to determine the probability of defective products and selection density. Fruit diameter is measured, survival rate and correction factor are calculated, and yield forecast is dynamically adjusted.
It enables precise quantitative assessment of the growth status of flowers and fruits, improves the accuracy and robustness of yield forecasting, allows for early identification of yield risks, rational allocation of resources, control of survey costs, and ensures the representativeness and practicality of the assessment.
Smart Images

Figure CN123070387A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of agricultural product yield prediction technology, and in particular to a method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits. Background Technology
[0002] In the production management of fruit trees and other flowering and fruiting agricultural products, accurate and timely forecasting of expected yields is crucial for agricultural scheduling, market sales, and supply chain management. However, in practical applications, it is impossible to dynamically reflect key dynamic changes such as physiological fruit drop and fruit enlargement throughout the entire development process of crops from flowering to maturity. It is also impossible to directly identify and quantify reproductive organs such as flowers and young fruits that directly determine yield. Furthermore, it lacks a reflection of the growth mechanism of flowers and fruits and fails to deeply integrate the real and continuous observation of the growth status of flowers and fruits with the progressive causal logic of yield formation. It is also impossible to dynamically and quantitatively calculate calibrable expected yields in a data-driven manner at each phenological stage from flowering to harvest.
[0003] Chinese Patent Application Publication No. CN120046809A describes an invention that uses image processing technology to determine the growth stage of crops in crop images from various sampling areas of farmland, matches the specific nutrient requirements corresponding to each crop growth stage, evaluates the influence coefficient of meteorological data on crop nutrient requirements by acquiring farmland meteorological data, generates a spatial distribution map of specific soil nutrients through spatial interpolation technology, generates a variable fertilization prescription map based on the influence coefficient of meteorological data on crop nutrient requirements and the specific nutrient requirements corresponding to each crop growth stage, and constructs a time series model based on the crop maturity time and yield of each historical crop cycle to obtain the current crop maturity time and yield prediction results and determine whether to issue an early warning. This invention can improve agricultural production efficiency and enhance agricultural risk management capabilities.
[0004] However, the existing technology still has the following problems: it does not take into account the monitoring of crop growth status, lacks the quantification of visual characteristics of the entire process of flowering and fruiting from flowering to maturity and the cross-phenological data coupling mechanism, which leads to the disconnect between yield prediction and actual growth status, and makes it impossible to achieve accurate yield quantification assessment. Summary of the Invention
[0005] To address this, the present invention provides a method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits. This method overcomes the problems in existing technologies that fail to consider monitoring crop growth status, lack quantification of visual characteristics of flowers and fruits throughout the entire process from flowering to maturity, and lack cross-phenological data coupling mechanisms, resulting in a disconnect between yield prediction and actual growth status, and making it impossible to achieve accurate yield quantitative assessment.
[0006] To achieve the above objectives, the present invention provides a method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits, comprising: Images of orchards during full bloom were acquired and color space segmentation was performed to obtain the initial number of flowers and the number of pollinated anthers, and the sampling strategy for flowering and fruiting trees was determined. Acquire images of young fruits during the young fruit stage and perform Hough circle transform detection to identify competitive weak fruits. Then, perform grayscale processing on the young fruit images to identify fruits at the branch tips. Finally, determine the effective base number based on the total number of young fruits, the competitive weak fruits, and the fruits at the branch tips. Extract the pixel features of each young fruit to determine the probability of defective fruit for each young fruit, and determine the selection density of flowering and fruiting trees based on the probability of defective fruit. Based on the sampling strategy and the selected density, a number of flowering and fruiting trees were selected in the orchard during the fruit enlargement period, and the transverse and longitudinal diameters of the fruits on each of the flowering and fruiting trees were collected to determine the average fruit weight. Acquire images of fruits during the fruiting period and perform texture feature analysis to determine whether each fruit is damaged and determine the fruit base number. Based on the effective base number and the fruit base number, determine the fruit survival rate. The correction factor is determined based on the initial number of flowers, the effective number of flowers, and the survival rate; the initial estimated quantity of the product is determined based on the number of fruits and the average fruit weight. The initial estimate of the product is adjusted based on the correction factor to obtain the expected output.
[0007] Furthermore, the process of determining the sampling strategy for flowering and fruiting trees includes: The initial flower base number is determined based on the number of connected components in several flower regions, wherein several flower regions are obtained by comparing and segmenting the pixel regions within the range of the hue and saturation of the orchard image with the hue of the reference flower. Extract the geometric center region of each flower as the stamen region in the flower region, and calculate the pixel ratio of the red channel to the blue channel. Flowers whose pixel ratio falls within the anther browning ratio range are identified as pollinated flowers, and the ratio of the number of pollinated flowers to the initial flower base is used as the anther browning ratio. The temperature data before and after the full bloom period are obtained and the effective accumulated temperature is calculated. The initial number of flowers is corrected by the anther color change ratio and the effective accumulated temperature to obtain the effective number of flowers. Based on the effective flower base, the sampling strategy is determined to be either a standard sampling strategy that selects the number of historical standards or a production reduction early warning strategy that reduces the number of selected flowers.
[0008] Furthermore, the process of identifying competitively weak results includes: Perform Hough circle transform detection on the young fruit image to obtain all first circular targets, and use the first circular targets as candidate young fruits; A spatial region is constructed with each candidate young fruit as the center and a preset radius is used. The number of candidate young fruits in the spatial region is counted, and the spatial region with a number of candidate young fruits greater than the benchmark number is marked as a dense competitive cluster. Candidate young fruits within each densely competitive cluster are sorted from largest to smallest according to their corresponding pixel radii. Candidate young fruits whose ranking is smaller than the preset last ranking are identified as the weakest competitive fruits and are removed.
[0009] Furthermore, the process of determining the effective base includes: The image of the young fruit is converted to grayscale. An edge detection algorithm is used to extract the branch edges and convert them into single-pixel skeleton lines. The endpoints of the single-pixel skeleton lines are detected as the branch end positions. Candidate young fruits whose path length between the center position and any of the branch end positions is less than the reference path length are determined as branch end fruits. The effective base number is the difference obtained by subtracting the number of competitively weak fruits and the number of fruits at the end of branches from the total number of candidate young fruits.
[0010] Furthermore, the process of determining the probability of defective products includes: For each candidate young fruit retained after Hough circle transform detection, its pixel diameter, branch-to-leaf ratio in local areas, and illumination intensity in local areas are extracted. The growth environment feature vector of each candidate young fruit is constructed using the pixel diameter, the branch-to-leaf ratio, and the light intensity. The growth environment feature vector is input into a preset defect probability prediction model to output the defect probability of the corresponding young fruit being judged as defective at the final harvest.
[0011] Furthermore, the process of determining the selection density of flowering and fruiting trees includes: The probability of defective fruit in all candidate young fruits in the orchard is aggregated, the orchard is divided into several sample areas according to a spatial grid, and the average of the probability of defective fruit in each sample area is calculated as the estimated defective rate. Based on the spatial distribution of the estimated defect rate, the orchard is divided into several risk level zones, and a differentiated selection density of flowering and fruit trees is set for different risk level zones. The higher the estimated defect rate, the greater the selection density of flowering and fruit trees within the corresponding risk level area.
[0012] Furthermore, the process of determining the average fruit weight includes: During the fruit enlargement period, select flowering and fruiting trees according to the corresponding selection quantity in each risk level area, and perform stratified sampling of each flowering and fruiting tree according to the canopy spatial layer, and select a number of fruits from each canopy spatial layer; The transverse and longitudinal diameters of each fruit were collected separately, and the square root of the product of the transverse and longitudinal diameters of each fruit was determined as the equivalent diameter. The average equivalent diameter is determined based on the average of the equivalent diameters of all fruits. The average equivalent diameter is then substituted into a preset fruit diameter-single fruit weight calibration model to output the average fruit weight.
[0013] Furthermore, the process of determining whether each fruit is damaged includes: The fruit image is converted to the Lab color space, and the local binary mode texture features of each fruit candidate region are extracted. The statistical histogram of the LBP feature values of all pixels in each fruit candidate region is used as the texture descriptor of the fruit. Calculate the chi-square distance between the texture descriptor of each fruit and the preset healthy fruit texture template; Fruits whose chi-square distance is greater than the benchmark chi-square distance are marked as damaged fruits; Fruits whose chi-square distance is less than or equal to the benchmark chi-square distance are marked as undamaged fruits.
[0014] Furthermore, the process of determining the retention rate includes: Perform Hough circle transform detection on the fruit image to obtain all the second circular targets, and determine the total number of second circular targets as the total number of fruits; The number of undamaged fruits out of the total number of fruits is determined as the base number of fruits. The retention rate is determined based on the effective base number and the fruit-bearing base number.
[0015] Furthermore, the process of determining the expected yield of fruit and flower products includes: The ratio of the effective base number to the initial flower base number is determined as the fruit set rate; The fruit set deviation rate is determined based on the fruit set rate and the benchmark fruit set rate. The retention deviation rate is determined based on the retention rate and the benchmark retention rate; The correction factor is determined based on the fruit setting deviation rate and the retention deviation rate; The initial estimated quantity of the product is calculated based on the number of fruits and the average fruit weight. The initial estimate of the product is revised based on the correction factor to obtain the expected output.
[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: by analyzing images during the full bloom period to obtain the initial flower base and pollination status, and combining this with effective accumulated temperature correction to obtain the effective flower base and determine the sampling strategy, the invention uses hue saturation segmentation and the ratio of red and blue channel pixels in the stamen region to identify pollinated flowers, thus objectively quantifying the number of flowers and the pollination ratio. By correcting the flower base through effective accumulated temperature, the effective flower base is made to better reflect the actual fruit set situation. Based on the effective flower base, the invention dynamically switches between standard sampling strategies or yield reduction warning strategies, which can proactively identify yield risks in the early stages and then rationally allocate subsequent resources for selecting flowering and fruiting trees, thereby further improving the accuracy of the method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits.
[0017] Furthermore, this invention detects young fruits and constructs dense competitive clusters using Hough circle transform. Weak competitive fruits are eliminated by sorting them by pixel radius. Simultaneously, fruits with limited development at the branch ends are eliminated using skeleton line endpoint detection, obtaining an effective base number. Young fruits naturally eliminated are eliminated from both spatial competition and location dimensions, ensuring that the retained effective base number more accurately reflects the final fruiting situation. A growth environment feature vector is constructed by extracting the pixel diameter, branch-to-leaf ratio, and light intensity of the retained young fruits. A defect probability prediction model outputs the defect probability of each young fruit to divide the orchard into different risk level areas and differentiate the selection density of flowering and fruiting trees. This ensures a positive match between sampling density and quality risk, controlling survey costs while guaranteeing the representativeness of assessments in different quality areas, thereby further improving the accuracy of the quantitative assessment method for expected agricultural yield based on the growth status of flowering and fruiting trees.
[0018] Furthermore, this invention employs stratified sampling based on differentiated density in risk areas during the fruit enlargement period, measures the transverse and longitudinal diameters of the fruit to calculate the equivalent diameter, and substitutes this into a fruit diameter-single fruit weight calibration model to obtain the average fruit weight, thus achieving objective quantification of single fruit weight. During the fruiting period, images are converted to the Lab color space and local binary pattern texture features are extracted. Damaged fruits are marked by comparing the chi-square distance with healthy fruit templates to determine the healthy fruit base, and the survival rate is calculated by combining the effective base. Texture feature analysis improves the consistency and accuracy of damaged fruit identification, providing reliable relevant parameters for the fruiting period for yield estimation throughout the entire growth period, thereby further improving the accuracy of the quantitative assessment method for expected agricultural product yield based on flower and fruit growth status.
[0019] Furthermore, this invention calculates the fruit set deviation rate and retention deviation rate by using the initial flower base number, effective base number, and retention rate, and generates a correction factor. It dynamically corrects the initial estimate obtained from the fruit set base number and average fruit weight to obtain the expected yield. It links the quantitative indicators of multiple stages, including flowers, young fruits, and fruit set, into a complete evaluation chain. By using the fruit set deviation rate and retention deviation rate to adjust the yield estimate, it effectively compensates for systematic deviations caused by flower and fruit drop, environmental fluctuations, etc., making the expected yield output closer to the actual harvest level. This improves the robustness and practicality of the evaluation method, thereby further improving the accuracy of the quantitative evaluation method for expected yield of agricultural products based on the growth status of flowers and fruits. Attached Figure Description
[0020] Figure 1 This is a flowchart of a method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits, according to an embodiment of the present invention. Figure 2 A logic decision diagram for determining the sampling strategy for flowering and fruiting trees in an embodiment of the present invention; Figure 3 This is a logic diagram for determining whether a candidate young fruit is a fruit at the end of a branch, according to an embodiment of the present invention. Figure 4 This is a logic diagram for determining whether each fruit is damaged, as shown in an embodiment of the present invention. Detailed Implementation
[0021] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.
[0022] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.
[0023] Please see Figure 1 The flowchart shown is a method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits, according to an embodiment of the present invention. The method in this embodiment includes: Step S1: Obtain orchard images during full bloom and perform color space segmentation to obtain the initial number of flowers and the number of pollinated anthers, and determine the sampling strategy for flowering and fruiting trees. Step S2: Obtain images of young fruits during the young fruit stage and perform Hough circle transform detection to identify competitive weak fruits. Perform grayscale processing on the young fruit images to identify fruits at the branch tips. Determine the effective base number based on the total number of young fruits, competitive weak fruits, and fruits at the branch tips. Step S3: Extract the pixel features of each young fruit to determine the probability of defective fruit for each young fruit, and determine the selection density of flowering and fruiting trees based on the probability of defective fruit. Step S4: Based on the sampling strategy and selection density, select a number of flowering and fruiting trees in the orchard during the fruit enlargement period, and collect the transverse and longitudinal diameters of the fruits on each flowering and fruiting tree to determine the average fruit weight. Step S5: Obtain images of fruits during the fruiting period and perform texture feature analysis to determine whether each fruit is damaged and determine the fruit base number. Determine the fruit survival rate based on the effective base number and the fruit base number. Step S6: Determine the correction factor based on the initial number of flowers, the number of flowers with effect, and the survival rate; determine the initial estimated quantity of product based on the number of fruits and the average fruit weight. Step S7: Adjust the initial product estimate based on the correction factor to obtain the expected output.
[0024] Please see Figure 2 As shown, this is a logic decision diagram for determining the sampling strategy for fruit trees in an embodiment of the present invention. The process of determining the sampling strategy for fruit trees includes: The initial flower base number is determined based on the number of connected components in several flower regions, wherein several flower regions are obtained by comparing and segmenting the pixel regions within the range of the baseline flower color tone based on the hue and saturation of the orchard image. Extract the geometric center region of each flower in the flower region as the stamen region, and calculate the pixel ratio of the red channel to the blue channel. Flowers with pixel ratios within the anther browning ratio range are identified as pollinated flowers, and the ratio of the number of pollinated flowers to the initial flower base is used as the anther browning ratio. Acquire temperature data before and after the peak flowering period and calculate the effective accumulated temperature. Correct the initial flower base number with the anther color change ratio and the effective accumulated temperature to obtain the effective flower base number. The sampling strategy is determined based on the effective flower base number, either by a standard sampling strategy that selects the number of flowers according to historical standards or by a production reduction early warning strategy that reduces the number of flowers selected.
[0025] In this embodiment, if the effective flower base number is greater than the baseline effective flower base number, the sampling strategy is determined to be the standard sampling strategy. If the effective flower base is less than or equal to the baseline effective flower base, then the sampling strategy is determined to be a production reduction early warning strategy.
[0026] Specifically, when determining the baseline effective flower base, we first review the orchard's production records for the past few years and extract the effective flower base calculated using the same method during the peak flowering period each year, along with the actual yield of the corresponding year. We then divide each year into normal yield years and reduced yield years based on yield levels. In normal yield years, the yield reaches or exceeds a certain percentage of the target yield for the corresponding variety, such as over 80%. In reduced yield years, the yield is significantly lower than the normal level. We list the effective flower base for normal yield years and the effective flower base for reduced yield years separately. The effective flower base for normal yield years will be concentrated in a higher range, while that for reduced yield years will be concentrated in a lower range. There is a transition zone between the two. We find the minimum effective flower base in normal yield years and the maximum effective flower base in reduced yield years. The area between the minimum and maximum values is the candidate range for the baseline effective flower base.
[0027] Specifically, the baseline flower tone range is the criterion for color segmentation. The baseline flower tone range requires the collection of color images of typical flower samples during the peak blooming period. During the peak blooming period, close-up images of flowers and complete images with backgrounds need to be collected from multiple points and angles under different lighting conditions. All the collected complete images are uniformly converted to HSV or HSI color space. From the pure petal pixels of the complete image, the distribution histograms of hue and saturation components are statistically analyzed to determine the concentrated area of flower pixels. The method for determining the concentrated area is as follows: first, find the center point of the joint distribution, that is, the mean of hue and saturation; then, according to the set confidence level, delineate a rectangular boundary that can contain the pixels of all complete images under the confidence level. The upper and lower limits of hue and saturation corresponding to the rectangular boundary constitute the baseline flower tone range.
[0028] Specifically, the anther browning ratio range is used to determine whether a flower has been pollinated. To determine the anther browning ratio range, firstly, images of pollinated flowers are collected. For each flower image, the pistil region is extracted, and the mean values of the red and blue channels of all pixels within the pistil region are calculated. The ratio of these two values is then taken as the anther color ratio of the flower. The above method is used to calculate the anther color ratio data of all pollinated flower samples one by one, resulting in a set of anther color ratio data for pollinated flowers.
[0029] The anther browning ratio range is determined based on the lower and upper quantiles of the statistical distribution of the ratios of pollinated flowers. For example, the fifth percentile of the distribution of the ratios of pollinated flowers is used as the lower limit of the anther browning ratio range, and the ninety-fifth percentile is used as the upper limit of the anther browning ratio range. Flowers whose anther color ratio falls within the anther browning ratio range are considered to have been pollinated.
[0030] Specifically, the effective accumulated temperature is used to correct the initial flower count obtained solely from image counting. The effective accumulated temperature can determine the degree of heat satisfaction required for flower organ development and pollination and fertilization. Insufficient or excessive effective accumulated temperature can lead to fruitless flowering and increased flower drop. Preferably, daily temperature data, including the highest and lowest temperatures, are obtained from one week before to one week after the peak flowering period. The recognized lower and upper biological limits for fruit tree varieties are selected. During the calculation, the average temperature values of the highest and lowest temperatures of the day are calculated first, and then the lower biological limit temperature is subtracted from the average temperature value. If the average temperature value of the day is lower than the lower biological limit temperature, the effective temperature value of the day is set to 0. If the average temperature value of the day is higher than the upper biological limit temperature, the upper biological limit temperature is used in the calculation. Several effective temperatures from one week before to one week after the peak flowering period are summed to obtain the effective accumulated temperature.
[0031] Specifically, the effective flower base number is the product of the initial flower base number and the pollination deviation rate, multiplied by the accumulated temperature deviation rate. The pollination deviation rate is the ratio of the anther color change ratio to the baseline anther color change ratio, which is the average of several anther color change ratios in historical data. The accumulated temperature deviation rate is the ratio of the effective accumulated temperature to the standard accumulated temperature, which is the pre-determined optimal effective accumulated temperature for the peak flowering period. The closer the effective accumulated temperature is to the standard accumulated temperature, the better the temperature conditions. The less the actual accumulated temperature is than the standard accumulated temperature, the worse the fruit set.
[0032] It is understandable that when the effective accumulated temperature is higher than the standard accumulated temperature, the accumulated temperature deviation rate is set as the upper limit to prevent overestimation of production due to short-term high temperatures, because high temperatures beyond the optimal temperature will only bring stress rather than promotion.
[0033] In this embodiment, the process of determining the weaker competing fruit includes: Perform Hough circle transform detection on the young fruit image to obtain all first circular targets, and use the first circular targets as candidate young fruits; A spatial region is constructed with each candidate young fruit as the center and according to a preset radius. The number of candidate young fruits in the spatial region is counted, and the spatial region with more than the baseline number of candidate young fruits is marked as a dense competitive cluster. Spatial regions with a number of candidate young fruits less than or equal to the baseline number are marked as non-dense competitive clusters; The candidate young fruits in each dense competitive cluster are sorted from largest to smallest according to their corresponding pixel radii. Candidate young fruits whose sorting is smaller than the preset last sorting position are identified as weak competitive fruits and are removed. Candidate young fruits whose sorting order is greater than or equal to the preset last sorting order are identified as non-competitive weak fruits and are retained.
[0034] Specifically, in the Hough circle transform detection of young fruit images, the minimum radius and the maximum radius together define the radius search range for the Hough circle transform detection. The minimum radius corresponds to the pixel size of the smallest resolvable young fruit in the image, and the maximum radius corresponds to the pixel size of the largest young fruit in the image.
[0035] Specifically, the accumulator threshold is the voting threshold used to determine whether a circular target is identified as a candidate young fruit. In the Hough circle transform, edge points vote for possible center positions, and the accumulator records the number of votes received by each candidate center. A higher accumulator threshold results in fewer detected circular targets but higher accuracy, eliminating non-circular interference areas; a lower threshold results in more sensitive detection but may introduce more false detections. The accumulator threshold can be determined through small-sample testing. Several typical young fruit images are selected, and the true young fruits are manually labeled. Then, different accumulator thresholds are iterated through, and the detection accuracy is calculated for each. The threshold with the highest accuracy is selected as the accumulator threshold for that variety under those shooting conditions.
[0036] It should be noted that the size of young fruits varies significantly among different varieties, and the radius range needs to be set according to the characteristics of the variety. The operation method is as follows: collect image samples of young fruits of this variety at a standard shooting distance, manually measure the pixel diameter of several typical young fruits in the image, take the minimum and maximum values, divide them by two respectively, and use them as the minimum and maximum radius parameters of the Hough circle transform. At the same time, the radius range can be appropriately widened by a certain proportion to cover individual differences.
[0037] When the pixel density or shooting distance of the image acquisition device changes, the number of pixels occupied by a young fruit of the same physical size in the image will change accordingly. Therefore, the radius parameter needs to be linked to the image resolution and the standard shooting distance. If the device is changed or the shooting distance is adjusted, the above calibration process should be repeated to update the radius range parameter and ensure consistent detection performance. Specifically, a reference object of known size is placed when acquiring images to establish a conversion relationship between physical size and pixel size, converting the physical fruit diameter parameter of the variety into the pixel radius range at the corresponding image resolution.
[0038] Specifically, the preset radius refers to the distance threshold used when generating a circular spatial region for evaluating the intensity of local competition, with the center of each candidate young fruit as the center. The radius of the circular spatial region is the preset radius. Specifically, young fruit images of the corresponding variety under normal fruit setting conditions can be collected, young fruit groups with obvious competitive relationships can be manually identified, the pixel distance distribution between the center points of the young fruit groups can be measured, and the mean of the pixel distance distribution can be taken as the preset radius.
[0039] Specifically, the baseline quantity is the threshold for determining whether a spatial area constitutes a densely competitive cluster. The baseline quantity needs to be determined based on the flowering and fruit setting characteristics of different fruit tree varieties and cultivation management experience. In practice, images of the young fruit stage of the corresponding variety in years with normal yields can be collected, and the distribution pattern of the number of fruits within a single branch group can be statistically analyzed. A critical value that can clearly distinguish between normal fruit setting and excessive fruit setting can be selected. For example, the number of fruits in each flower cluster area can be counted, and the upper limit of single cluster fruit setting that maximizes the yield can be taken as the baseline quantity.
[0040] Specifically, the preset last ranking is the selection ranking of the weakest fruits in the competition. The preset last ranking needs to be determined in combination with the optimal number of fruits to be retained in each fruiting part of the corresponding variety under the target cultivation mode. The number of fruits that are finally retained in the densely fruiting parts can be extracted from the historical high-yield and high-quality orchard data to determine the number of dominant fruits. The preset last ranking is set to this number plus one. For example, if a variety retains one central fruit in an inflorescence, the preset last ranking is two, that is, only the dominant fruit ranked first is retained; if two fruits are retained, the preset last ranking is three.
[0041] Please see Figure 3 As shown, this is a logic diagram for determining whether a candidate young fruit is a fruit at the end of a branch, according to an embodiment of the present invention. The process of determining the effective base includes: The young fruit image is converted to grayscale. An edge detection algorithm is used to extract the branch edges and convert them into single-pixel skeleton lines. The endpoints of the single-pixel skeleton lines are detected as the branch end positions. Candidate young fruits whose path length between the center position and any branch end position is less than the baseline path length are determined as branch end fruits. Candidate young fruits whose path lengths at the center and at all branch ends are greater than or equal to the baseline path length are classified as non-branch end fruits. The effective base number is the difference obtained by subtracting the number of weaker competing fruits from the total number of candidate young fruits, and then subtracting the number of fruits at the end of branches.
[0042] Specifically, the baseline path length is the threshold for determining whether a candidate young fruit is a terminal fruit of a branch. Images of young fruits under normal management conditions for the corresponding variety are collected, and young fruits that are clearly located at the end of the branch and non-terminal young fruits located in the middle and lower part of the branch are manually marked. The path length from the center of these two types of young fruits to the nearest branch skeleton line endpoint is measured to obtain the path length distribution of terminal fruits and the path length distribution of non-terminal fruits. The upper limit of the path length distribution of terminal fruits, that is, the maximum value of the path length in terminal fruits, is taken as the baseline path length.
[0043] Understandably, the baseline path length should be adapted to the natural internode length of the branch tip section, generally not exceeding the length of two to three internodes at the tip of the corresponding variety's fruiting branch, in order to ensure that young fruits that are close to the tip and have poor development potential can be accurately identified within a reasonable range.
[0044] In this embodiment, the process of determining the probability of defective products includes: For each candidate young fruit retained after Hough circle transform detection, its pixel diameter, branch-to-leaf ratio in local areas, and illumination intensity in local areas are extracted. The growth environment feature vectors of each candidate young fruit are constructed using pixel diameter, branch-to-leaf ratio, and light intensity. The growth environment feature vector is input into a pre-set defect probability prediction model to output the probability of the corresponding young fruit being judged as defective at the final harvest.
[0045] Specifically, the branch-to-leaf ratio is the ratio of the area of green pixels to the area of fruit pixels in a local region, and the light intensity is the average HSV brightness of the local region.
[0046] Specifically, a pre-defined defect probability prediction model is used to transform the local growth environment characteristics of young fruit into a quantifiable defect risk indicator. A higher defect probability indicates poorer development conditions and a greater risk of defection. Specifically, a logistic regression model is chosen as the basic architecture for the defect probability prediction model. The logistic regression model consists of a linear layer and a non-linear activation function. The linear layer performs a weighted summation of the three input features—pixel diameter, branch-to-leaf ratio, and light intensity—and adds a bias term to obtain a linear prediction value. This linear prediction value is mapped to the zero-to-one interval using a log-probability function, and the output is the probability that the young fruit will be judged as defective at harvest.
[0047] Throughout the orchard's historical production cycle, at least one thousand candidate young fruits were collected during the young fruit stage, collecting data on pixel diameter, branch-to-leaf ratio, and light intensity. Each young fruit was tagged and tracked, and its actual quality assessment at harvest was recorded, indicating whether it was a substandard fruit, thus forming a tagged feature dataset. The ratio of substandard to non-substandard samples in the dataset should roughly correspond to the perennial substandard rate for that variety. When the naturally collected sample ratio is unbalanced, oversampling or undersampling methods can be used to adjust it, ensuring a near-balance between the two types of samples. The dataset was then randomly divided into a training set and a validation set at a ratio of seven to three.
[0048] The logistic regression model was trained using the training set with the following hyperparameters: L2 regularization was selected, the reciprocal of the regularization strength was set to 1.0, LBFGS was selected as the solver, and the maximum number of iterations was set to 500. The training process aimed to minimize the cross-entropy loss function, and the weight coefficients and bias terms of each feature in the linear layer were determined through iterative optimization.
[0049] The trained model is evaluated using a validation set. The primary validation metric is the area under the receiver operating characteristic curve (AUC), which measures the model's overall ability to distinguish between defective and non-defective products. A validation AUC of 0.75 or higher is considered sufficient to meet application requirements. Simultaneously, the model's output probabilities are calibrated to ensure that the average error between the output probability and the actual defect rate does not exceed five percentage points. Once these metrics are met, the weights and biases of the logistic regression model are permanently stored and used as a pre-defined defect probability prediction model.
[0050] Specifically, the growth environment feature vector is input into a preset grading probability prediction model. The grading probability prediction model outputs the grading probability of each size grade in several size grades corresponding to the final harvest of the young fruit. The set of all grading probabilities is used as the probability vector of size grading. The probability vectors of all candidate young fruits in the orchard are aggregated, the sum of probabilities corresponding to each size grade is calculated, and the ratio of this sum to the total number of candidate young fruits is calculated to obtain the estimated proportion of each size grade.
[0051] In this embodiment, the graded probability prediction model is used to transform the individual development information of young fruit into the ability to predict the future fruit size distribution. This model needs to be trained in advance through supervised learning. First, based on the market circulation standards for the corresponding fruit varieties, the finally harvested fruits are divided into four size grades according to their individual fruit weight, and each grade is labeled with a corresponding individual fruit weight range: premium grade fruits are those weighing 300 grams or more; first grade fruits are those weighing between 200 and 300 grams; second grade fruits are those weighing between 150 and 200 grams; and inferior grade fruits are those weighing less than 150 grams. This forms a clear four-category labeling system, with each young fruit corresponding to a unique size grade label at harvest.
[0052] A multi-class logistic regression model was chosen as the basic architecture for the grading probability prediction model. The model input is a feature vector of the candidate young fruit's growth environment, including three features: pixel diameter, branch-to-leaf ratio, and light intensity. The model linearly combines the input features, mapping the linear prediction values to probability values corresponding to the four size grades through a normalized exponential function, with the sum of all probability values being one. The model output is a four-dimensional probability vector, with each dimension corresponding to the grading probability of a size grade.
[0053] Throughout the historical production cycle, at least two thousand candidate young fruits were collected during the young fruit stage, their growth environment feature vectors were uniquely identified, and tracked until harvest. At harvest, the weight of each fruit was measured, and they were categorized into the corresponding size class according to the aforementioned labeling system, forming a labeled dataset with paired feature vectors and size class labels. The number of samples in each size class of the dataset should be as balanced as possible; if the number of samples in some classes is too small, data augmentation can be used to supplement them.
[0054] The labeled dataset was randomly divided into training and validation sets in a 7:3 ratio. The following hyperparameters were configured for training: L2 regularization was selected, the reciprocal of the regularization strength was set to 1.0, LBFGS was selected as the solver, and the maximum number of iterations was set to 500. The training process aimed to minimize the multi-class cross-entropy loss function, and the weight coefficients and bias terms corresponding to each feature of the model were determined through iterative optimization.
[0055] The trained model is evaluated using a validation set. The primary validation metric is multi-class accuracy, which is the proportion of samples whose predicted size class matches the actual size class out of the total number of samples in the validation set. When the validation set accuracy reaches 70% or higher, the model's predictive ability is considered to meet the application requirements. Simultaneously, the probability outputs for each level are calibrated and validated to ensure that the deviation between the model's output probabilities and the actual occurrence rates is within an acceptable range. After meeting the above metrics, the weight parameters of the multi-class logistic regression model are permanently stored and used as a pre-built hierarchical probability prediction model.
[0056] In this embodiment, the process of determining the selection density of flowering and fruiting trees includes: The probability of defective fruit in all candidate young fruits in the orchard is aggregated, the orchard is divided into several sample areas according to spatial grid, and the average of the probability of defective fruit in each sample area is calculated as the estimated defective rate. Based on the spatial distribution of the estimated defect rate, the orchard is divided into several risk level zones, and a differentiated selection density of flowering and fruit trees is set for different risk level zones. Among them, the higher the estimated defect rate, the greater the selection density of flowering and fruit trees in the corresponding risk level area.
[0057] Specifically, risk-level zones are different management blocks into which orchards are divided based on the spatial distribution of defective fruit probabilities. This allows for differentiated sampling and monitoring. First, the orchard is spatially gridded, divided into several sample areas using grids of fixed sizes. The grid side length can be set according to the orchard area and young fruit density, for example, using five or ten meters, to ensure that each sample area contains a sufficient number of candidate young fruits for statistical analysis. Second, the estimated defective fruit rate for each sample area is calculated. The previously obtained defective fruit probabilities for each candidate young fruit are categorized according to their corresponding sample areas. Within the same sample area, all... The average probability of defective fruit among candidate young fruits is used as the estimated defect rate. Then, based on the numerical distribution of the estimated defect rate in each sample area within the orchard, a threshold is set to divide the orchard into three risk level zones: a low-risk zone (estimated defect rate below the lower threshold), a medium-risk zone (estimated defect rate between the lower and higher thresholds), and a high-risk zone (estimated defect rate above the higher threshold). The lower and higher thresholds can be selected based on the distribution characteristics of the estimated defect rate in the orchard, for example, the third quantile and the second quantile can be used as the lower and higher thresholds, respectively.
[0058] Specifically, selection density refers to the number of flowering and fruiting trees whose fruit diameters need to be measured within each fixed number of trees selected in each risk level area during the fruit enlargement period. First, a baseline selection density is determined, which is applicable to the area in the orchard with the lowest risk and the most stable fruit development. Specifically, based on the mean and standard deviation of the fruit diameter in normal years of the corresponding variety, within a preset allowable error, such as ±5%, the minimum required number of sample trees is calculated using a statistical sample size formula as the baseline selection density.
[0059] Specifically, the selected density of different risk level areas is used as a benchmark, and the selected density is multiplied by different magnification factors. The higher the risk level, the larger the magnification factor. The variance of the estimated defect rate in each risk level area is calculated. The variance of the low-risk area is used as a benchmark, and the ratio of the variance of other areas to the variance of the low-risk area is used as the magnification factor for the corresponding risk level area.
[0060] In this embodiment, the process of determining the average fruit weight includes: During the fruit enlargement period, select flowering and fruiting trees according to the corresponding selection quantity in each risk level area, and perform stratified sampling of each flowering and fruiting tree according to the canopy spatial layer, and select a number of fruits from each canopy spatial layer; Collect the transverse and longitudinal diameters of each fruit separately, and determine the square root of the product of the transverse and longitudinal diameters of each fruit as the equivalent diameter. The average equivalent diameter is determined based on the average of the equivalent diameters of all fruits. The average equivalent diameter is then substituted into the preset fruit diameter-single fruit weight calibration model to output the average fruit weight.
[0061] Specifically, the pre-set fruit diameter-single fruit weight calibration model is used to convert the size parameters of the fruit into an estimated value of the single fruit weight. The pre-set fruit diameter-single fruit weight calibration model needs to be established in advance through destructive sampling calibration, and it is an empirical regression model.
[0062] Specifically, during the historical production cycle, fruits are harvested in batches and by canopy layer from the fruit enlargement stage to maturity, covering the entire size range from the smallest to the largest. Immediately after harvesting, the transverse and longitudinal diameters of each fruit are measured using image measurement tools, and its equivalent diameter is calculated using the same method. Subsequently, the actual fresh weight of each fruit is weighed using an electronic balance, forming a calibration dataset consisting of pairs of equivalent diameters and corresponding single fruit weights. Since fruit weight is proportional to the cube of the diameter, and the relationship between fruit diameter and fruit weight exhibits a power function relationship, a power function form can be adopted, i.e., fruit weight equals the coefficient multiplied by a power of the equivalent diameter, as the mathematical structure of the model. Using the equivalent diameter in the calibration dataset as the independent variable and the single fruit weight as the dependent variable, the coefficients and exponential parameters for the corresponding varieties are obtained through least squares fitting, yielding the complete expression of the fruit diameter-single fruit weight calibration model.
[0063] Specifically, during the expansion period survey of the current season, sample fruits are harvested at differentiated densities, and their transverse and longitudinal diameters are measured. The equivalent diameter of each fruit is calculated using the same method. The arithmetic mean of the equivalent diameters of all fruits is calculated to obtain the average equivalent diameter of the current season. The average equivalent diameter is then substituted into the established power function formula. At this point, the coefficients and exponents in the power function formula are known constants, and the independent variable is the average equivalent diameter. After substituting into the formula, the function output value is the average fruit weight of the current season.
[0064] Please see Figure 4 As shown, this is a logic diagram for determining whether each fruit is damaged in an embodiment of the present invention. The process of determining whether each fruit is damaged includes: The fruit images are converted to the Lab color space, and the local binary mode texture features of each fruit candidate region are extracted. The statistical histogram of the LBP feature values of all pixels in each fruit candidate region is used as the texture descriptor of the fruit. Calculate the chi-square distance between the texture descriptor of each fruit and the preset healthy fruit texture template; Fruits with a chi-square distance greater than the baseline chi-square distance were marked as damaged fruits. Fruits with a chi-square distance less than or equal to the baseline chi-square distance are marked as undamaged fruits.
[0065] Specifically, the pre-set healthy fruit texture template serves as a reference for determining whether there is surface damage on the fruit. It needs to be constructed in advance by collecting images of healthy, undamaged fruits before the fruiting period. First, several normally growing fruits with no visible defects are selected in the orchard. Clear images of their intact surfaces are taken under natural light conditions, covering fruit surfaces from different orientations and lighting angles to enhance the template's universality. Second, each clear image is converted to the Lab color space, and the fruit region containing only the fruit surface is extracted through color segmentation. For each clear image's fruit region, local binary pattern feature values are calculated pixel-by-pixel. The frequency of LBP feature values for all pixels within this region is statistically analyzed to construct a texture histogram for each clear image. The LBP operator can use a basic 3x3 neighborhood uniform mode to capture local texture details while maintaining appropriate dimensionality. Finally, the texture histograms of all healthy fruits are normalized and averaged; that is, the frequency average of all fruits is taken for each texture histogram to obtain a standardized statistical histogram, which is the pre-set healthy fruit texture template.
[0066] Specifically, the baseline chi-square distance is the threshold for determining whether a fruit is damaged. The smaller the chi-square distance, the closer the texture is to a healthy state; the larger the chi-square distance, the higher the degree of texture abnormality. The baseline chi-square distance needs to be calibrated by combining the chi-square distance distribution of healthy fruits and known damaged fruits. First, a batch of fruit images is selected, including both those confirmed as healthy by manual inspection and those confirmed to have various typical types of damage. For each fruit image, the candidate regions of the fruit are extracted and texture descriptors are calculated according to the same process. The chi-square distance between each image and the pre-set healthy fruit texture template is then calculated, and the chi-square distance distribution is statistically analyzed. The baseline chi-square distance can be set as the upper quantile of the chi-square distance distribution of healthy fruit samples, such as the 95th or 99th percentile. The specific value of the quantile can be adjusted according to the tolerance for missed and false positives of damaged fruits in actual management. The higher the quantile, the stricter the judgment of damaged fruits and the lower the missed judgment rate.
[0067] In this embodiment, the process of determining the retention rate includes: Perform Hough circle transform detection on the fruit image to obtain all second circular targets, and determine the total number of second circular targets as the total number of fruits; The number of undamaged fruits in the total number of fruits is determined as the base number of fruits; The retention rate is determined based on the effective base number and the fruit-bearing base number.
[0068] Specifically, the retention rate is the ratio of the number of plants bearing fruit to the number of plants with effective results.
[0069] In this embodiment, the process of determining the expected yield of fruit and flower agricultural products includes: The ratio of the effective number of flowers to the initial number of flowers is determined as the fruit set rate; Determine the fruit set deviation rate based on the fruit set rate and the benchmark fruit set rate; The retention deviation rate is determined based on the retention rate and the baseline retention rate; The correction factor was determined based on the fruit setting deviation rate and the retention deviation rate. The initial estimated product quantity is calculated based on the base number of fruits and the average fruit weight. The initial product estimate is adjusted based on the correction factor to obtain the expected output.
[0070] Specifically, the expected yield is the product of the correction factor and the initial product estimate, which is the product of the number of fruits on the tree and the average fruit weight.
[0071] Specifically, the fruit set deviation rate is the difference between the fruit set rate and the benchmark fruit set rate, calculated by ratio to the benchmark fruit set rate; the retention deviation rate is the difference between the retention rate and the benchmark retention rate, calculated by ratio to the benchmark retention rate.
[0072] Specifically, a positive fruit set deviation rate or retention deviation rate indicates that the actual performance of this stage is better than the benchmark level; a negative fruit set deviation rate or retention deviation rate indicates that the actual performance of this stage is worse than the benchmark level; and a zero fruit set deviation rate or retention deviation rate indicates that the actual performance of this stage is on par with the benchmark level.
[0073] Specifically, the fruit setting deviation rate and the retention deviation rate are compared, and the smaller of the two is taken. The sum of one and the smaller value is determined as the correction factor. That is, the correction factor is equal to one plus the smaller of the fruit setting deviation rate and the retention deviation rate.
[0074] Specifically, when both the fruit setting deviation rate and the retention deviation rate are positive, both stages are better than the benchmark. The smaller advantage is added by one as a correction factor, and if the correction factor is greater than one, the expected yield is moderately increased. When both are negative, both stages are worse than the benchmark. The larger disadvantage (i.e., the larger absolute value but smaller numerical value) is added by one, and if the correction factor is less than one, the expected yield is revised downward, and the magnitude of the revision is determined by the stage most severely affected. When one is positive and the other is negative, the excess performance of one stage cannot fully compensate for the absolute loss of the other stage. The negative value is added by one, and if the correction factor is less than one, the expected yield is revised downward. When both deviation rates are zero, the correction factor is one, and the expected yield is consistent with the initial estimate.
[0075] Specifically, when determining the benchmark fruit setting rate and benchmark survival rate, historical data of the same variety from the past 3 to 5 years are collected from the orchard. Data anomalies caused by extreme disasters are removed, and the average of the remaining years' fruit setting rate and survival rate are calculated as the benchmark fruit setting rate and benchmark survival rate, respectively.
[0076] It should be noted that if the orchard lacks historical data for more than three consecutive years, the regional standard value of the corresponding variety will be used as the benchmark fruit setting rate and benchmark survival rate when this method is first applied. After three years of application, the method will automatically switch to statistical values based on the orchard's own historical data.
[0077] The final evaluation report output by this embodiment includes quantitative evaluation results in three dimensions: the estimated proportion of premium-grade fruit, the probability of substandard fruit, and the expected yield.
[0078] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.
Claims
1. A method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits, characterized in that, include: Images of orchards during full bloom were acquired and color space segmentation was performed to obtain the initial number of flowers and the number of pollinated anthers, and the sampling strategy for flowering and fruiting trees was determined. Acquire images of young fruits during the young fruit stage and perform Hough circle transform detection to identify competitive weak fruits. Then, perform grayscale processing on the young fruit images to identify fruits at the branch tips. Finally, determine the effective base number based on the total number of young fruits, the competitive weak fruits, and the fruits at the branch tips. Extract the pixel features of each young fruit to determine the probability of defective fruit for each young fruit, and determine the selection density of flowering and fruiting trees based on the probability of defective fruit. Based on the sampling strategy and the selected density, a number of flowering and fruiting trees were selected in the orchard during the fruit enlargement period, and the transverse and longitudinal diameters of the fruits on each of the flowering and fruiting trees were collected to determine the average fruit weight. Acquire images of fruits during the fruiting period and perform texture feature analysis to determine whether each fruit is damaged and determine the fruit base number. Based on the effective base number and the fruit base number, determine the fruit survival rate. The correction factor is determined based on the initial number of flowers, the effective number of flowers, and the survival rate; the initial estimated quantity of the product is determined based on the number of fruits and the average fruit weight. The initial estimate of the product is adjusted based on the correction factor to obtain the expected output.
2. The method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits according to claim 1, characterized in that, The process of determining the sampling strategy for flowering and fruiting trees includes: The initial flower base number is determined based on the number of connected components in several flower regions, wherein several flower regions are obtained by comparing and segmenting the pixel regions within the range of the hue and saturation of the orchard image with the hue of the reference flower. Extract the geometric center region of each flower as the stamen region in the flower region, and calculate the pixel ratio of the red channel to the blue channel. Flowers whose pixel ratio falls within the anther browning ratio range are identified as pollinated flowers, and the ratio of the number of pollinated flowers to the initial flower base is used as the anther browning ratio. The temperature data before and after the full bloom period are obtained and the effective accumulated temperature is calculated. The initial number of flowers is corrected by the anther color change ratio and the effective accumulated temperature to obtain the effective number of flowers. Based on the effective flower base, the sampling strategy is determined to be either a standard sampling strategy that selects the number of historical standards or a production reduction early warning strategy that reduces the number of selected flowers.
3. The method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits according to claim 2, characterized in that, The process of identifying competitive disadvantages includes: Perform Hough circle transform detection on the young fruit image to obtain all first circular targets, and use the first circular targets as candidate young fruits; A spatial region is constructed with each candidate young fruit as the center and a preset radius is used. The number of candidate young fruits in the spatial region is counted, and the spatial region with a number of candidate young fruits greater than the benchmark number is marked as a dense competitive cluster. Candidate young fruits within each densely competitive cluster are sorted from largest to smallest according to their corresponding pixel radii. Candidate young fruits whose ranking is smaller than the preset last ranking are identified as the weakest competitive fruits and are removed.
4. The method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits according to claim 3, characterized in that, The process of determining the effective base includes: The image of the young fruit is converted to grayscale. An edge detection algorithm is used to extract the branch edges and convert them into single-pixel skeleton lines. The endpoints of the single-pixel skeleton lines are detected as the branch end positions. Candidate young fruits whose path length between the center position and any of the branch end positions is less than the reference path length are determined as branch end fruits. The effective base number is the difference obtained by subtracting the number of competitively weak fruits and the number of fruits at the end of branches from the total number of candidate young fruits.
5. The method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits according to claim 4, characterized in that, The process of determining the probability of defective products includes: For each candidate young fruit retained after Hough circle transform detection, its pixel diameter, branch-to-leaf ratio in local areas, and illumination intensity in local areas are extracted. The growth environment feature vector of each candidate young fruit is constructed using the pixel diameter, the branch-to-leaf ratio, and the light intensity. The growth environment feature vector is input into a preset defect probability prediction model to output the defect probability of the corresponding young fruit being judged as defective at the final harvest.
6. The method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits according to claim 5, characterized in that, The process of determining the selection density of flowering and fruiting trees includes: The probability of defective fruit in all candidate young fruits in the orchard is aggregated, the orchard is divided into several sample areas according to a spatial grid, and the average of the probability of defective fruit in each sample area is calculated as the estimated defective rate. Based on the spatial distribution of the estimated defect rate, the orchard is divided into several risk level zones, and a differentiated selection density of flowering and fruit trees is set for different risk level zones. The higher the estimated defect rate, the greater the selection density of flowering and fruit trees within the corresponding risk level area.
7. The method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits according to claim 6, characterized in that, The process of determining the average fruit weight includes: During the fruit enlargement period, select flowering and fruiting trees according to the corresponding selection quantity in each risk level area, and perform stratified sampling of each flowering and fruiting tree according to the canopy spatial layer, and select a number of fruits from each canopy spatial layer; The transverse and longitudinal diameters of each fruit were collected separately, and the square root of the product of the transverse and longitudinal diameters of each fruit was determined as the equivalent diameter. The average equivalent diameter is determined based on the average of the equivalent diameters of all fruits. The average equivalent diameter is then substituted into a preset fruit diameter-single fruit weight calibration model to output the average fruit weight.
8. The method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits according to claim 7, characterized in that, The process of determining whether each fruit is damaged includes: The fruit image is converted to the Lab color space, and the local binary mode texture features of each fruit candidate region are extracted. The statistical histogram of the LBP feature values of all pixels in each fruit candidate region is used as the texture descriptor of the fruit. Calculate the chi-square distance between the texture descriptor of each fruit and the preset healthy fruit texture template; Fruits whose chi-square distance is greater than the benchmark chi-square distance are marked as damaged fruits; Fruits whose chi-square distance is less than or equal to the benchmark chi-square distance are marked as undamaged fruits.
9. The method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits according to claim 8, characterized in that, The process of determining the retention rate includes: Perform Hough circle transform detection on the fruit image to obtain all the second circular targets, and determine the total number of second circular targets as the total number of fruits; The number of undamaged fruits out of the total number of fruits is determined as the base number of fruits. The retention rate is determined based on the effective base number and the fruit-bearing base number.
10. The method for quantitatively assessing the expected yield of agricultural products based on the growth status of flowers and fruits according to claim 9, characterized in that, The process of determining the expected yield of fruit and flower crops includes: The ratio of the effective base number to the initial flower base number is determined as the fruit set rate; The fruit set deviation rate is determined based on the fruit set rate and the benchmark fruit set rate. The retention deviation rate is determined based on the retention rate and the benchmark retention rate; The correction factor is determined based on the fruit setting deviation rate and the retention deviation rate; The initial estimated quantity of the product is calculated based on the number of fruits and the average fruit weight. The initial estimate of the product is revised based on the correction factor to obtain the expected output.
Citation Information
Patent Citations
Agricultural industry chain information data monitoring, processing and analyzing method
CN120046809A
