High-quality fruit production potential assessment method based on citrus tree canopy scale

Abstract

The invention discloses a high-quality fruit production potential assessment method based on a citrus tree canopy scale, which belongs to the technical field of remote sensing and comprises the following steps of field data acquisition and detection, unmanned aerial vehicle multispectral image acquisition and processing, citrus tree canopy information extraction and high-quality fruit production potential assessment. Wherein the high-quality fruit production potential evaluation comprises sample fruit tree division, evaluation index selection, evaluation index threshold division, evaluation model construction and evaluation result precision inspection. On the basis of the unmanned aerial vehicle multispectral remote sensing technology, high-quality fruit production potential evaluation is carried out on the citrus fruit trees, the citrus orchard production potential is pre-judged in advance, the problems that current high-quality fruit production potential evaluation is large in difficulty, low in efficiency, lack of quantitation and the like are solved, high-quality fruit accurate information is provided for fruit farmers and fruit merchants, and the method is suitable for popularization and application. The method helps to formulate a reasonable harvesting plan and a market pricing strategy, improves the competitiveness of high-quality citrus fruits, and provides a technical support for promoting the development of the citrus industry.

Description

Technical Field

[0001] The present invention belongs to the technical field of remote sensing, and particularly relates to a method for evaluating the production potential of high-quality fruits based on the canopy scale of citrus fruit trees. Background Art

[0002] The evaluation of the production potential of high-quality fruits in orchards refers to the comprehensive analysis and quantitative evaluation of the quality and yield potential of fruits in orchards through scientific methods. This evaluation usually combines environmental factors, fruit tree characteristics, management measures, and modern technical means (such as remote sensing, sensors, model analysis, etc.), aiming to provide a scientific basis for orchard management, resource optimization, and market decision-making.

[0003] In the pre-listing period of citrus, mastering the production potential of high-quality fruits of citrus fruit trees is crucial for fruit farmers' sales price evaluation, sales plans, fruit tree management, etc. Therefore, it is necessary to carry out the evaluation of the production potential of high-quality fruits in orchards to provide accurate information on high-quality fruits for fruit farmers and fruit merchants.

[0004] Unmanned aerial vehicle (UAV) multi-spectral remote sensing has the characteristics of low cost, lightness, speed, and rich spectral information, and has been widely used in large-scale agricultural information monitoring. The existing technology can estimate the citrus yield through UAV multi-spectral remote sensing. For example, the article "Research on Citrus Yield Estimation by Fusing UAV Images and Machine Learning" published in Transactions of the Chinese Society for Agricultural Machinery, authors: Wu Lifeng, Xu Wenhao, Pei Qingbao. This research plan obtained remote sensing image data of citrus fruit maturity period through a DJI multi-spectral UAV, and extracted visible light and multi-spectral band indices from the images as feature variables, and respectively constructed classification models for the presence or absence of citrus fruits, and estimation models for the number and quality of fruits, achieving good estimation results, providing a new method for the rapid estimation of citrus orchard yield.

[0005] However, in the existing technology, no method for evaluating the production potential of high-quality fruits through remote sensing technology has been found. Therefore, based on the UAV multi-spectral remote sensing technology, the present invention proposes a method for evaluating the production potential of high-quality fruits based on the canopy scale of citrus fruit trees, to solve the problems of large evaluation difficulty, low efficiency, and lack of quantification of the current production potential of high-quality fruits, provide accurate information on high-quality fruits for fruit farmers and fruit merchants, help formulate reasonable harvesting plans and market pricing strategies, so as to improve the competitiveness of high-quality citrus fruits and provide technical support for promoting the development of the citrus industry. Summary of the Invention

[0006] The present invention provides a method for evaluating the production potential of high-quality fruits based on the canopy scale of citrus fruit trees, to solve the problems of large evaluation difficulty, low efficiency, and lack of quantification of the current production potential of high-quality fruits.

[0007] To achieve the above object, the technical solution adopted by the present invention is:

[0008] A method for evaluating the production potential of high-quality fruits based on the canopy scale of citrus fruit trees, including:

[0009] S1. Field data collection and detection: Conduct field data collection in the test area, calibrate multiple sample fruit trees with different production potentials, and collect sample fruit tree data, including: fruit diameter, fruit quantity, fruit color, peel blemish rate, and fruit tree growth; Collect the fruits of the sample fruit trees for the following detections: soluble sugar, vitamin C, soluble solids, titratable acid, and edible rate, so as to obtain sample fruit detection data;

[0010] S2. Collect multi-spectral images of the test area by an unmanned aerial vehicle and process them to obtain orchard multi-spectral images and an unmanned aerial vehicle digital surface model (Digital Surface Model, DSM);

[0011] S3. According to the orchard multi-spectral images and DSM, extract the canopy information of citrus fruit trees, identify the canopy information of individual citrus fruit trees, and obtain the canopy data of each citrus fruit tree. The canopy data includes the canopy boundary and area of the fruit tree;

[0012] S4. Evaluate the production potential of high-quality fruits of citrus fruit trees, including the following steps:

[0013] S41. Sample fruit tree classification: According to the canopy data obtained in S3, the sample fruit tree data and sample fruit detection data obtained in S1, divide multiple sample fruit trees into three categories: sample trees of excellent grade in fruit production potential, sample trees of good grade in fruit production potential, and sample trees of general grade in fruit production potential;

[0014] S42. Selection of evaluation indicators: By calculating the correlations between various different vegetation indices and soluble sugar, vitamin C, soluble solids, titratable acid, and edible rate, select appropriate vegetation indices together with the fruit tree canopy area and canopy fruit area as evaluation indicators; The canopy fruit area can be extracted using the Normalized Pigment Chlorophyll Index (NPCI);

[0015] S43. Division of evaluation indicator thresholds: According to the selected evaluation indicators, divide the evaluation indicator thresholds in the order of excellent, good, and general grades, and set the membership degrees of the thresholds;

[0016] S44. Construction of an evaluation model: Based on the selected evaluation indicators, construct a production potential model of high-quality fruits to calculate the comprehensive index. The specific calculation formula is as follows:

[0017]

[0018] In the formula: y is the comprehensive index; t is the evaluation indicator weight (1 / number of evaluation indicators); Pi It is the membership degree of the i-th evaluation factor; i is the evaluation index number; n is the number of evaluation indicators; when y ≥ x1, it is classified into the high-quality level, when x2 < y < x1, it is classified into the good level, when y ≤ x2, it is classified into the general level, and x1 and x2 are the thresholds corresponding to the comprehensive index y;

[0019] S45. Check the accuracy of the evaluation results.

[0020] Furthermore, in S1, the field data collection also includes collecting: the row spacing of citrus fruit tree planting, the planting data of each row of fruit trees, and the information of missing plants.

[0021] Furthermore, S3 specifically includes the following steps:

[0022] S31. Orchard zoning based on terrain: According to the orchard multispectral image, the test area is divided into several zones based on the DSM value;

[0023] S32. Extraction of fruit tree canopy information: Based on the regional boundaries of several zones, objects are formed using the multiscale segmentation algorithm, and a multi-region canopy information extraction model is constructed to achieve the extraction of fruit tree canopy information by region;

[0024] S33. Identification of single citrus fruit tree canopy information: According to the trunk position information of the first fruit tree in each row, the first trunk point is formed, and all trunk points are automatically generated according to the row spacing of citrus fruit tree planting; the center point of two trunk points is obtained, and it is automatically extended a m to the left and right and b m up and down to form a rectangular region of interest; where a and b are set values, a roughly encompasses the boundary region, and b is slightly larger than the canopy width, which can be set with reference to the canopy projection area; the DSM is cropped using the region of interest to produce a DSM data set, and the segmentation line of each DSM is determined through the watershed algorithm; the canopy information is segmented using the segmentation line to form the canopy boundary of single citrus fruit trees;

[0025] S34. Inspection of the accuracy of canopy information identification: Combining the planting data of each row of fruit trees and the information of missing plants collected in the field, check whether the identification of single citrus fruit tree canopy information is accurate, and manually correct the inaccurate places. The manual correction includes deleting the non-citrus fruit tree canopy.

[0026] Furthermore, in step S34, the manual correction also includes manually drawing the missing citrus fruit tree canopy and calculating the area of all canopies; evenly distributing and manually drawing the canopies of n fruit trees, calculating the canopy projection area as the measured value, and performing a correlation analysis with the projection area (extracted value) of the extracted canopy information of n fruit trees. The accuracy of the canopy area is checked using the coefficient of determination R 2 and the root mean square error RMSE.

[0027] Further, in S41, the indicators for classifying the sample trees into the high-quality grade of fruit production potential are as follows: the fruit tree canopy area is relatively large, the growth of the fruit tree is good, the fruit diameter is greater than c mm, the number of fruits is relatively large, the fruit shape is round, the color is uniform, the fruit skin has no or slight speckling, the soluble sugar and edible rate are high, and the titratable acid is low; the indicators for classifying the sample trees into the good grade of fruit production potential are as follows: the fruit tree canopy area is medium, the growth of the fruit tree is good, the fruit diameter is greater than d mm, the number of fruits is medium, the color is slightly uneven, the fruit skin speckling does not exceed e%, and the soluble sugar, edible rate, and titratable acid are medium; the indicators for classifying the sample trees into the general grade of fruit production potential are as follows: the fruit tree canopy area is relatively small, the growth of the fruit tree is average, the appearance is deformed, the fruit surface is blue, the fruit skin speckling exceeds e%, the soluble sugar is low, and the titratable acid is high; where c, d, and e are all set values.

[0028] Further, in S42, various different vegetation indices used to calculate the correlation are all closely related to chlorophyll, nitrogen and phosphorus elements, growth, and moisture.

[0029] Further, in S42, the selected evaluation indicators are NDVI, TCAVI, REOSAVI, the fruit tree canopy area, and the canopy fruit area; NDVI is the normalized difference vegetation index, TCAVI is the transformed chlorophyll absorption reflectance index, and REOSAVI is the red-edge optimized soil-adjusted vegetation index.

[0030] Further, in step S44, the comprehensive indices of the sample fruit trees are sorted in descending order. x1 is the comprehensive index corresponding to the f% of the sample fruit trees before sorting, and x2 is the comprehensive index corresponding to the g% of the sample fruit trees before sorting; where f and g are both set values.

[0031] Further, in step S45, the accuracy P, recall rate R, and F value in information retrieval and statistics are introduced for accuracy evaluation, and the calculation formulas are as follows:

[0032]

[0033] Among them, TP, FP, and FN respectively represent the number of fruit tree plants whose grades are correctly evaluated for the sample fruit trees, the number of fruit tree plants whose grades are wrongly evaluated, and the number of fruit tree plants that are omitted and not evaluated for the corresponding grades. The accuracy P represents the proportion of the number of fruit tree plants that are correctly evaluated among the fruit trees of a certain grade. The recall rate R represents the proportion of the number of fruit tree plants whose grades are correctly evaluated to the number of sample fruit tree plants of the corresponding grade. F is a comprehensive description of the accuracy and recall rate. When all the fruit trees in the orchard can be correctly evaluated for their grades, F = 100%; conversely, when all the sample fruit trees of all grades are wrongly classified, F = 0; the higher the F value, the better the evaluation result.

[0034] Due to the adoption of the above technical solution, the present invention has the following beneficial effects:

[0035] 1. The present invention is based on the multi - spectral remote sensing technology of unmanned aerial vehicles and ground measurement means, constructs an evaluation model for the production potential of high - quality citrus fruits at the canopy scale of citrus fruit trees, fills the gap in the lack of evaluation technology for the production potential of high - quality citrus fruits, solves the problems of difficult evaluation, low efficiency, and lack of quantification of the production potential of high - quality fruits, can effectively improve the competitiveness of high - quality citrus fruits, and provides technical support for promoting the development of the citrus industry.

[0036] 2. The present invention selects multiple vegetation indices with relatively high correlations as evaluation indicators, divides appropriate threshold ranges for the evaluation indicators, and combines ground measurement means. After multiple tests and corrections, it ensures the prediction accuracy of the evaluation model and realizes the accurate evaluation of the production potential of citrus fruit quality.

[0037] 3. The present invention can realize the early prediction of the production potential of citrus orchards, provide accurate information on high - quality fruits for fruit farmers and fruit merchants, help formulate reasonable harvesting plans and market pricing strategies, solve the problem of unequal acquisition prices for high - quality fruits, increase farmers' income, and is of great significance for the transformation and upgrading of modern agriculture.

[0038] 4. The present invention uses the low - altitude remote sensing technology of unmanned aerial vehicles to carry out the evaluation of the production potential of high - quality citrus fruits at the canopy scale of citrus fruit trees, provides a new application scenario for exploring the deep integration development of the low - altitude economy and industries, and plays a positive role in promoting the development of smart agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is the technical roadmap of the present invention;

[0040] Figure 2 is the multi - spectral image of the orchard obtained in S2 of the present invention;

[0041] Figure 3 is the schematic diagram of the experimental area division in S31 of the present invention;

[0042] Figure 4 is the effect diagram of the extraction of the canopy information of fruit trees in S32 of the present invention;

[0043] Figure 5 is the schematic diagram of the recognition process of the canopy information of a single citrus fruit tree in S33 of the present invention;

[0044] Figure 6 is the recognition result diagram of the canopy information of a single citrus fruit tree in S33 of the present invention;

[0045] Figure 7 is the accuracy test diagram of the recognition of the canopy information of a single citrus fruit tree in S34 of the present invention;

[0046] Figure 8 is the evaluation result of the production potential of citrus fruits in S44 of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts belong to the scope of protection of the present invention.

[0048] Embodiment 1

[0049] As Figure 1 shown, the present invention provides a method for evaluating the production potential of high-quality fruits based on the canopy scale of citrus fruit trees, including S1 - S4.

[0050] S1. Field data collection and detection: Conduct field data collection in the test area. The collected data includes: the row spacing of citrus fruit tree planting, the planting data of fruit trees in each row, and the missing plant information. Calibrate multiple sample fruit trees with different production potentials, and collect sample fruit tree data, including: fruit diameter, fruit quantity, fruit color, peel blemish rate, and fruit tree growth. Collect the fruits of the sample fruit trees for the following detections: soluble sugar, vitamin C, soluble solids, titratable acid, and edible rate, so as to obtain the sample fruit detection data.

[0051] S2. Collect the multispectral images of the test area by an unmanned aerial vehicle (UAV) and process them to obtain the orchard multispectral images and the UAV digital surface model (DSM). The present invention uses the DJI Phantom 4 multispectral imaging to obtain high-definition data of citrus orchards in Wuming District. The collection and processing of UAV multispectral images are roughly divided into three parts: data collection, data production, and band recombination.

[0052] 1) Multispectral data collection: The collection of multispectral data is mainly carried out by using the aerial survey UAV of the DJI Phantom 4 multispectral imaging system. The multispectral imaging system integrates 6 multispectral cameras, which are respectively responsible for visible light imaging and multispectral imaging. All cameras have a resolution of 2 million pixels and a global shutter, and can collect high-precision multispectral data to help complete agricultural monitoring and environmental monitoring work efficiently.

[0053] 2) Multispectral data production: Import the multispectral photo data into the Pix4d software. The software automatically identifies the camera information on the photo data and groups the photos according to six band lenses: visible light, red, green, blue, near-infrared, and red edge. After the data import is completed, select the agricultural multispectral data processing template in the Pix4d software to enter the processing interface. Under the processing option index calculator, calibrate the data of each band lens. Taking the reflectance panel placed in the test area (reflectance is 50%) as the reference, use the reflectance panel pixel information collected by each lens to perform corresponding calibration processing settings. Finally, check the image data to be exported to complete the data production.

[0054] 3) Band data recombination: Import the data of each band into the Arcgis software, and perform band recombination through the Composite bands function in the ArcToolbox toolbox. The obtained multispectral image of the orchard in the test area is as Figure 2 shown, and the image resolution is 0.03m.

[0055] S3. According to the multispectral image of the orchard and the DSM, extract the canopy information of citrus fruit trees, identify the canopy information of individual citrus fruit trees, and obtain the canopy data of each citrus fruit tree. The canopy data includes the boundary and area of the fruit tree canopy. S3 specifically includes the following steps:

[0056] S31. Orchard zoning based on terrain: According to the multispectral image of the orchard, divide the test area into several zones based on the DSM value. In this embodiment, the test area has a gentle slope terrain, regular fruit tree planting, equipped with automatic irrigation devices, good management level, lush fruit tree growth, a continuous strip-shaped distribution of the canopy, unclear boundaries of individual fruit tree canopies, and overgrown weeds around the fruit trees. When constructing a classification model for the entire park, misclassification is likely to occur. To accurately extract the fruit tree canopy, divide the test area into 4 zones according to the DSM value. The zoning situation is as Figure 3 shown.

[0057] S32. Fruit tree canopy information extraction: Based on the regional boundaries of several zones, use the multiscale segmentation algorithm to form objects and construct a multi-region canopy information extraction model to achieve the extraction of fruit tree canopy information by region.

[0058] Among them, the citrus orchard can be divided into vegetation information and non-vegetation information (such as roads, rocks, bare land, etc.). Use the Normalized Difference Vegetation Index (NDVI) to distinguish between vegetation and non-vegetation information. Weeds around the fruit trees are likely to interfere with the canopy information. Considering the characteristic that there is a certain height difference between the fruit tree canopy and the weeds, use the drone DSM to distinguish the surface weeds and the fruit canopy information.

[0059] In this embodiment, taking the orchard multi-spectral image as the object, based on the boundary of the partition area, the multi-scale segmentation (scale 500) algorithm is used to form objects, and a multi-region canopy information extraction model is constructed, as shown in Table 1 specifically. The extraction of the fruit tree canopy information for each region is realized. Through a small amount of manual correction, the output results are merged, and the canopy information extraction results are as follows Figure 4 shown

[0060] Partition number Extraction model Partition 1 NDVI≥ - 0.1 and DSM≥110.5 Partition 2 NDVI≥0.05 Partition 3 NDVI≥0 and DSM≥111 Partition 4 NDVI≥0 and DSM≥109

[0061] Table 1. Multi-region canopy information extraction model

[0062] S33. Identification of the canopy information of a single citrus fruit tree: According to the trunk position information of the first fruit tree in each row, the first trunk point is formed, and all trunk points are automatically generated according to the planting row spacing of citrus fruit trees. The center point of two trunk points is obtained, and it is automatically extended outward by a m on both the left and right sides and by b m upward and downward to form a rectangular region of interest; where a and b are set values, a roughly encompasses the dividing line area, and b is slightly larger than the canopy width, and can be set with reference to the canopy projection area. The DSM is cropped using the region of interest to produce a DSM data set, and the dividing line of each DSM is determined through the watershed algorithm. The canopy information is segmented using the dividing line to form the canopy information of a single citrus fruit tree

[0063] In this embodiment, the citrus fruit trees in the test area are planted relatively regularly. The average row spacing between adjacent fruit tree trunks is 4.5 m, the trunk spacing in each row of fruit trees is between 1.4 and 1.6 m, and the average trunk spacing is 1.5 m. The trunk position information of the first fruit tree in each row is obtained through field measurement means to form the first trunk point, and all trunk points are automatically generated at a distance of 1.5 m. The center point of two trunk points is obtained, and it is automatically extended outward by 0.25 m on both the left and right sides and by 3 m upward and downward to form a rectangular region of interest. The DSM is cropped using the region of interest to produce a DSM data set, and the dividing line of each DSM is determined through the watershed algorithm. The canopy information is segmented using the dividing line to form the canopy information of a single fruit tree. The identification process and extraction results of the canopy information of a single fruit tree are respectively as follows Figure 5 、 6 shown

[0064] S34. Inspection of the identification accuracy of canopy information: Combining the planting data and missing plant information of each row of fruit trees collected in the field, it is inspected whether the identification of the canopy information of a single citrus fruit tree is accurate, and the inaccurate places are manually corrected. Among them, manual correction includes deleting the canopy of non-citrus fruit trees, and also includes manually drawing the missing canopy of citrus fruit trees, and calculating the area of all canopies; n fruit tree canopies are manually drawn evenly distributed, and the canopy projection area is calculated as the measured value, and a correlation analysis is carried out with the projection area (extracted value) of the extracted canopy information of n fruit trees. Using the determination coefficient R 2The root mean square error RMSE is used to test the accuracy of the canopy area.

[0065] In this embodiment, through the identification of the canopy information of single citrus fruit trees, a total of 3,411 citrus plants were identified. Based on the field-recorded location information of the cut plants, the canopy information was inspected row by row. It was found that there were 36 misidentifications of the canopy, and the identification accuracy was 98.95%. Due to the cutting of plants, the weeds grew lushly, and the branches of adjacent fruit trees covered each other, resulting in misclassification of the canopy. After processing the non-citrus canopy information, 3,375 single-tree canopies of fruit trees were identified.

[0066] To test the accuracy of the extraction of single-tree canopies, the number of citrus fruit trees planted in each row was sampled and surveyed in the left and right areas along the road. The comparison between the actual measured number of planted fruit trees and the identified number is shown in Table 2 below. The overall accuracy of the identification of single-tree canopies of fruit trees is 98.27%.

[0067] Table 2 shows that except for the 14th row in the right area where the identified number is more than the actual measured plants, the rest are mainly missed classifications. The main reason is that the planting spacing of fruit trees in the test area is not uniform, the fruit trees are planted densely, the branches grow lushly, there are many phenomena of continuous canopies, and the branches and leaves at the tree edge are piled up, which raises the edge elevation and causes displacement of the boundary line.

[0068] To test the accuracy of the canopy area, 17 fruit tree canopies were hand-drawn evenly distributed, and the projected area of the canopy was calculated as the measured value. A correlation analysis was carried out with the projected area (extracted value) of the 17 canopy information extracted, as Figure 7 shown, the coefficient of determination R 2 is 0.869, and the root mean square error RMSE is 0.489 m 2 , indicating that the extraction accuracy of the crown width projected area is good.

[0069]

[0070]

[0071] Table 2. Comparison table of the actual measured number and the monitored number of citrus plants (unit: plant)

[0072] S4. Evaluate the production potential of high-quality fruits of citrus fruit trees, including S41 - S44.

[0073] S41. Sample fruit tree classification: According to the canopy data obtained from S3, the sample fruit tree data obtained from S1, and the sample fruit detection data, multiple sample fruit trees are divided into three categories: sample trees with excellent fruit production potential grade, sample trees with good fruit production potential grade, and sample trees with general fruit production potential grade.

[0074] The indicators for classifying sample trees into the high-quality grade of fruit production potential are as follows: the fruit tree canopy area is relatively large, the growth of the fruit tree is relatively good, the fruit diameter is greater than c mm, the number of fruits is relatively large, the fruit shape is round, the color is uniform, the fruit skin has no or slight blemishes, and the soluble sugar and edible rate are high while the titratable acid is low.

[0075] The indicators for classifying sample trees into the good grade of fruit production potential are as follows: the fruit tree canopy area is medium, the growth of the fruit tree is good, the fruit diameter is greater than d mm, the number of fruits is medium, the color is slightly uneven, the blemished area of the fruit skin does not exceed e%, and the soluble sugar, edible rate, and titratable acid are medium.

[0076] The indicators for classifying sample trees into the general grade of fruit production potential are as follows: the fruit tree canopy area is relatively small, the growth of the fruit tree is general, the appearance is deformed, the fruit surface is blue, the blemished area of the fruit skin exceeds e%, and the soluble sugar is low while the titratable acid is high.

[0077] Among them, c, d, and e are all set values. In this embodiment, c = 65, d = 60, and e = 30. Whether the fruit tree canopy area is relatively large, medium, or relatively small, whether the growth of the fruit tree is relatively good, good, or general, and whether the number of fruits is relatively large, medium, or general, etc., can be divided by setting thresholds according to the actual situation, like the "fruit diameter division", which will not be elaborated in detail in this article.

[0078] S42. Selection of evaluation indicators: Calculate the correlations between various different vegetation indices and soluble sugar, vitamin C, soluble solids, titratable acid, and edible rate. Among them, the various different vegetation indices used for calculating the correlations are all closely related to chlorophyll, nitrogen and phosphorus elements, growth, and moisture. Select appropriate vegetation indices together with the fruit tree canopy area and the canopy fruit area as evaluation indicators. The canopy fruit area can be extracted using the Normalized Pigment Chlorophyll Index (NPCI).

[0079] When calculating the correlation, the correlation coefficient is represented by r, and its value range is [-1, 1]. r > 0 indicates a positive correlation, r < 0 indicates a negative correlation. Generally, |r| < 0.5 indicates a weak correlation, 0.5 ≤ |r| < 0.8 indicates a moderate correlation, and |r| ≥ 0.8 indicates a high correlation. In this embodiment, 30 vegetation indices are selected for correlation calculation, as shown in Table 3 for details.

[0080]

[0081]

[0082]

[0083] Table 3. Details of Vegetation Indices

[0084] The specific results of the calculation of the vegetation index correlation are shown in Table 4 below. Table 4 shows that 30 vegetation indexes are weakly correlated with soluble sugar and vitamin C, mSR705 is moderately correlated with soluble solids (r is 0.566), TCAVI is moderately correlated with titratable acid (r is 0.707), and REOSAVI is highly negatively correlated with edible rate (r is -0.889). Therefore, TCAVI and REOSAVI are respectively selected as the representative indexes for titratable acid and edible rate.

[0085]

[0086] Table 4 Correlation Coefficient Calculation Table

[0087] In addition, NDVI can effectively reflect the crop health status, and this index is selected as the evaluation index for the growth status of citrus fruit trees; the canopy area and canopy fruit area of citrus fruit trees reflect the growth status and the number of fruits, and the canopy fruit area can be extracted by using the NPCI index. Therefore, the evaluation indexes selected in this embodiment are NDVI, TCAVI, REOSAVI, the canopy area of fruit trees, and the canopy fruit area, and the fruit production potential is evaluated through these 5 indexes.

[0088] S43. Divide the evaluation index threshold: According to the selected evaluation indexes, divide the evaluation index threshold in the order of excellent, good, and general levels, and set the membership degree of the threshold.

[0089] In this embodiment, taking the fruit tree canopy as a unit, the average values of NDVI, TCAVI, and REOSAVI are obtained, and the maximum values, minimum values, and average values of the 5 evaluation indexes of the sample fruit trees (10 trees of each grade sample tree) and all fruit trees in the test area are respectively counted. The evaluation index threshold is divided in the order of excellent, good, and general levels. The threshold division process is as follows: (1) First, consider the excellent level. With a span of the average value ±0.05 (which can be appropriately increased or decreased according to the actual situation), count the number of trees in different ranges from the average value to the maximum and minimum values respectively. When the cumulative sum of the number of trees exceeds 50% of the excellent level sample trees, this range value is the excellent level threshold. (2) Count the number of good level sample trees falling into the excellent level threshold as k, and use 10 of the good level sample trees minus k to obtain the actual number n of the good level sample trees. Based on the excellent level threshold, gradually decrease or increase the range value, and at the same time calculate the number of fruit trees corresponding to the range value. When the number exceeds 50% of n, this range value is the good level range; (3) The range values outside the excellent and good level sample trees are the general level thresholds.

[0090] To solve the problem of inconsistent dimensions of the evaluation indexes, the membership degrees of the 5 evaluation indexes for the excellent, good, and general levels are respectively set to 0.5, 0.3, and 0.1, and the 5 evaluation indexes have the same weight, that is, the weight value t = 0.2.

[0091] S44. Evaluation model construction: Based on the selected evaluation indicators, a high-quality fruit production potential model is constructed to calculate the comprehensive index. The specific calculation formula is as follows:

[0092]

[0093] In the formula: y is the comprehensive index; P i is the membership degree of the i-th evaluation factor; i is the evaluation index number; n is the number of evaluation indicators. The higher the y value, the higher the possibility that the corresponding citrus fruit tree belongs to high-quality fruits, and vice versa.

[0094] Evaluation criteria: If y≥x1, it is classified as the high-quality level; if x2<y<x1, it is classified as the good level; if y≤x2, it is classified as the general level. x1 and x2 are the thresholds corresponding to the comprehensive index y. Specifically, the comprehensive indexes of the sample fruit trees are sorted in descending order. x1 is the comprehensive index corresponding to the f% of the sample fruit trees before sorting, and x2 is the comprehensive index corresponding to the g% of the sample fruit trees before sorting; where f and g are both set values. In this embodiment, f = 40 and g = 75.

[0095] In this embodiment, the evaluation index thresholds and the classification of the comprehensive index levels in the test area are shown in Table 5. If y≥0.36, it is classified as the high-quality level; if 0.24<y<0.36, it is classified as the good level; if y≤0.24, it is classified as the general level.

[0096]

[0097] Table 5. Evaluation index thresholds and comprehensive index level classification table

[0098] S45. Check the accuracy of the evaluation results. The accuracy P, recall rate R, and F value in information retrieval and statistics are introduced for accuracy evaluation. The calculation formulas are as follows:

[0099]

[0100] Among them, TP, FP, and FN respectively represent the number of fruit tree plants correctly evaluated at a certain level, the number of fruit tree plants mis-evaluated at a certain level, and the number of fruit tree plants omitted and not evaluated at the corresponding level. The accuracy P represents the proportion of the number of fruit tree plants correctly evaluated among the fruit tree plants of a certain level. The recall rate R represents the proportion of the number of fruit tree plants correctly evaluated at a certain level to the number of fruit tree plants of the corresponding level in the sample. F is a comprehensive description of the accuracy and recall rate. When all the fruit trees in the orchard can be correctly evaluated at a certain level, F = 100%; on the contrary, when all the fruit tree plants of all levels are misclassified, F = 0; the higher the F value, the better the evaluation result.

[0101] In this embodiment, 47 sample trees (30 were involved in the division of quality grade thresholds and 17 were not) were used to conduct the accuracy test of fruit production potential evaluation. The specific results are shown in Table 6 below. Table 6 shows that the sample fruit trees of high-quality, good, and average fruit production potential grades are 13, 19, and 15 respectively. The overall accuracy of the evaluation results is 85.11%. The evaluation accuracy of the high-quality grade is the best, followed by the average grade. Among them, the accuracy rate P of the high-quality grade is 85.71%, the recall rate R is 92.31, and the F value is 88.89%; the accuracy rate P of the good grade is 83.33%, the recall rate R is 78.95, and the F value is 81.08%; the accuracy rate P, recall rate R, and F value of the average grade are all 86.67%. Among the sample fruit trees of the high-quality grade, 1 was misclassified as the good grade; among the sample fruit trees of the good grade, 2 were misclassified as the high-quality grade and 2 were misclassified as the average grade; among the sample fruit trees of the average grade, 2 were misclassified as the good grade. Since the NDVI, TCAVI, and REOSAVI indicators are not linearly related, there is a certain degree of confusion among the three grades of fruit trees. The confusion between the good-grade fruit trees and the high-quality and average-grade fruit trees is the most obvious.

[0102]

[0103] Table 6. Accuracy Test Table of Fruit Production Potential Evaluation

[0104] In this embodiment, the fruit production potential of high-quality citrus fruit trees in the experimental orchard was evaluated. The evaluation results are as Figure 8 shown, where the high-quality, good, and average grade fruit trees are 1187, 1517, and 671 respectively, accounting for 35.17%, 44.95%, and 19.88% of the orchard fruit trees.

[0105] The above description is a detailed description of the preferred and feasible embodiments of the present invention. However, the embodiments are not intended to limit the scope of the patent application of the present invention. Any equivalent changes or modifications made under the technical spirit disclosed by the present invention shall fall within the scope of the patent covered by the present invention.

Claims

1. A method for evaluating the production potential of high-quality fruits based on the canopy scale of citrus trees, characterized in that: include: S1. Field data collection and testing: Field data collection was conducted in the experimental area, and multiple sample fruit trees with different production potentials were calibrated. The sample fruit tree data were collected, including: fruit diameter, fruit quantity, fruit color, fruit skin flower rate, and fruit tree growth; the fruits of the sample fruit trees were collected for the following tests: soluble sugar, vitamin C, soluble solids, titratable acid, and edible rate, and the sample fruit test data were obtained; S2. Collect multispectral images of the test area through drones and process them to obtain multispectral images of the orchard and drone digital surface model (DSM); S3. Extract the canopy information of citrus trees based on the orchard multispectral image and DSM, and identify the canopy information of individual citrus trees to obtain the canopy data of each citrus tree. The canopy data includes the canopy boundary and area of ​​the fruit tree. S4. Evaluate the production potential of high-quality citrus fruit trees, including the following steps: S41, classification of sample fruit trees: according to the canopy data obtained in S3 and the sample fruit tree data and sample fruit detection data obtained in S1, multiple sample fruit trees are classified into three categories: sample trees with excellent fruit production potential, sample trees with good fruit production potential, and sample trees with average fruit production potential; S42, selection of evaluation indexes: by calculating the correlation between various vegetation indices and soluble sugar, vitamin C, soluble solids, titratable acid, and edible rate, appropriate vegetation indexes are selected together with the canopy area of ​​fruit trees and the canopy fruit area as evaluation indicators; the canopy fruit area can be extracted using the Normalized Pigment Chlorophyll Index (NPCI); S43, dividing the evaluation index threshold: dividing the evaluation index threshold in the order of high-quality, good, and general grades according to the selected evaluation index, and setting the membership degree of the threshold; S44. Evaluation model construction: Based on the selected evaluation indicators, a high-quality fruit production potential model is constructed to calculate the comprehensive index. The specific calculation formula is as follows: Where: y is the comprehensive index; t is the evaluation index weight (1 / number of evaluation indicators); P i is the i-th evaluation factor membership; i is the evaluation index number; n is the number of evaluation indicators; y≥x1 is divided into high-quality level, x2<y<x1 is divided into good level, y≤x2 is divided into general level, x1 and x2 are the thresholds corresponding to the comprehensive index y; S45. Check the accuracy of the evaluation results.

2. The method for evaluating the production potential of high-quality fruit based on the canopy scale of citrus trees according to claim 1, characterized in that: In S1, field data collection also includes collecting: the spacing between citrus fruit tree rows, the planting data of each row of fruit trees, and information on missing trees.

3. The method for evaluating the production potential of high-quality fruit based on the canopy scale of citrus trees according to claim 2, characterized in that: S3 specifically includes the following steps: S31. Orchard zoning based on terrain: Based on the orchard multispectral imagery, the test area was divided into several zones according to the DSM values; S32, fruit tree canopy information extraction: Based on the regional boundaries of several partitions, a multi-scale segmentation algorithm is used to form objects, and a multi-region canopy information extraction model is constructed to realize regional fruit tree canopy information extraction; S33, single citrus tree canopy information identification: according to the trunk position information of the first fruit tree in each row, form the first trunk point, and automatically generate all trunk points according to the row spacing of citrus trees; obtain the center point of the two trunk points, automatically expand am to the left and right sides, and expand bm up and down to form a rectangular area of ​​interest; where a and b are set values, a roughly encompasses the boundary area, and b is slightly larger than the canopy width, which can be set with reference to the canopy projection area; use the area of ​​interest to crop the DSM, produce a DSM data set, and determine the segmentation line of each DSM through the watershed algorithm; use the segmentation line to segment the canopy information to form the canopy boundary of a single citrus tree; S34. Canopy information recognition accuracy test: Combined with the planting data of each row of fruit trees and the missing tree information collected in the field, check whether the canopy information of individual citrus trees is accurately identified, and make manual corrections to inaccurate identifications, including deleting non-citrus tree canopies.

4. The method for evaluating the production potential of high-quality fruit based on the canopy scale of citrus trees according to claim 3, characterized in that: In step S34, manual correction also includes manually drawing the missing citrus tree canopies and calculating the area of ​​all canopies; evenly distributing the canopies of n fruit trees manually, calculating the canopy projection area as the measurement value, and performing correlation analysis with the extracted canopy information projection area of ​​n trees (extracted value), and using the determination coefficient R 2 The accuracy of canopy area was tested by using the root mean square error (RMSE).

5. The method for evaluating the production potential of high-quality fruit based on the canopy scale of citrus trees according to claim 4, characterized in that: In S41, the indicators for classification of sample trees as high-quality grade with fruit production potential are: large canopy area, good growth, fruit diameter greater than c mm, large number of fruits, round fruit shape, uniform color, no or slight peel flower, high soluble sugar and edible rate, and low titratable acid; the indicators for classification of sample trees as good grade with fruit production potential are: medium canopy area, good growth, fruit diameter greater than d mm, medium number of fruits, slightly uneven color, peel flower not exceeding e%, medium soluble sugar, edible rate and titratable acid; the indicators for classification of sample trees as average grade with fruit production potential are: small canopy area, average growth, deformed appearance, green fruit surface, peel flower more than e%, low soluble sugar and high titratable acid; among them, c, d, and e are all set values.

6. The method for evaluating the production potential of high-quality fruit based on the canopy scale of citrus trees according to claim 1, characterized in that: In S42, the various vegetation indices used to calculate correlations are closely related to chlorophyll, nitrogen and phosphorus, growth potential, and water.

7. The method for evaluating the production potential of high-quality fruit based on the canopy scale of citrus trees according to claim 6, characterized in that: In S42, the selected evaluation indicators are NDVI, TCAVI, REOSAVI, fruit tree canopy area and canopy fruit area; NDVI is the normalized difference vegetation index, TCAVI is the converted chlorophyll absorption reflectance index, and REOSAVI is the red edge optimized soil conditioned vegetation index.

8. The method for evaluating the production potential of high-quality fruit based on the canopy scale of citrus trees according to claim 7, characterized in that: In step S44, the comprehensive indexes of the sample fruit trees are sorted in descending order, x1 is the comprehensive index corresponding to the first f% of the sample fruit trees in the sort, and x2 is the comprehensive index corresponding to the first g% of the sample fruit trees in the sort; wherein f and g are both set values.

9. The method for evaluating the production potential of high-quality fruit based on the canopy scale of citrus trees according to claim 8, characterized in that: In step S45, the accuracy P, recall R and F value in information retrieval and statistics are introduced to perform accuracy evaluation, and the calculation formula is as follows: Among them, TP, FP, and FN represent the number of sample fruit trees with correctly assessed grades, the number of fruit trees with incorrectly assessed grades, and the number of fruit trees that were omitted and not assessed at the corresponding grade, respectively; the precision rate P represents the proportion of fruit trees with correctly assessed grades in fruit trees of a certain grade; the recall rate R represents the proportion of fruit trees with correctly assessed grades to the number of sample fruit trees of the corresponding grade; F is a comprehensive description of the precision rate and recall rate. When all fruit trees in the orchard can be correctly assessed, F = 100%; otherwise, when all grade sample fruit trees are incorrectly graded, F = 0.

Images