An oil-tea camellia fruit tree yield prediction system, method and storage medium based on a three-dimensional laser scanner

Abstract

The present invention provides an oil-tea camellia fruit yield prediction system, method and storage medium based on a three-dimensional laser scanner, relating to the technical field of oil-tea camellia fruit yield prediction, including: in the data acquisition stage, using a three-dimensional laser scanner and a hyperspectral camera to obtain fruit volume, geometric feature data and internal component content data, laying a solid foundation for subsequent analysis. By calculating the actual density of the fruit, a density correction index is generated, fully considering multiple factors affecting fruit density, effectively improving the calculation accuracy of fruit quality. A quality prediction model is built based on a deep learning network, which can capture the complex relationship between fruit quality and various factors. The present invention greatly enhances the adaptability to different growth environments and fruit individuals, significantly improves the generalization ability of the model. Multiple modules work together to construct a complete system for yield prediction. Each link is closely connected, reducing manual intervention and data errors, and greatly improving the work efficiency of oil-tea camellia fruit yield detection.

Description

Technical Field

[0001] The present invention relates to the technical field of oil-tea camellia fruit yield prediction, and specifically to an oil-tea camellia fruit yield prediction system, method and storage medium based on a three-dimensional laser scanner. Background Technique

[0002] In recent years, although deep learning has achieved good results in object detection, it often requires complex operation models and a large amount of data training. Moreover, due to the lush branches and leaves in the fruit tree canopy and severe occlusion, when extracting the number of fruits by taking digital images, it is difficult for the images to include all the fruits on the fruit tree, which will lead to a large error between the finally calculated number of fruits and the actual number, and further cause a greater error in the calculation of fruit yield. With the rapid development of information technology, three-dimensional laser scanning technology has become increasingly mature. This technology can collect point cloud data from different angles through lidar, comprehensively obtain the three-dimensional point cloud information of oil-tea camellia trees, and is not affected by factors such as object color and environmental light, with good robustness. By processing and analyzing these point cloud data, fruits can be distinguished. Compared with manual measurement and image-based methods, the detection speed is faster and the accuracy is higher, providing a new effective way for the intelligent measurement of oil-tea camellia fruit yield.

[0003] In the prior art, an oil-tea camellia fruit yield detection method provided by the publication number CN112215184B includes the following steps: obtaining three-dimensional point cloud data of an oil-tea camellia tree through a three-dimensional laser scanner, and preprocessing the collected three-dimensional point cloud data to obtain a three-dimensional point cloud data set of the oil-tea camellia tree; segmenting the three-dimensional point cloud data set of the oil-tea camellia tree to obtain a three-dimensional point cloud data set containing leaf and fruit information; performing density statistics on the three-dimensional point cloud data set of the leaf and fruit information, setting a threshold, filtering out the leaves, and obtaining N three-dimensional point cloud data subsets of fruits, where N is a positive integer; calculating the fruit radius corresponding to each fruit in the fruit radius array according to the three-dimensional point cloud data subsets of the fruits; calculating the fruit mass corresponding to each fruit radius in the fruit radius array according to the basic parameters in the relationship model between the fruit radius and the fruit mass, and accumulating to calculate the yield of the whole fruit tree.

[0004] However, there are still the following deficiencies. From the above statement, it can be seen that the prior art only obtains the fruit radius by means of a three-dimensional laser scanner, fails to collect the geometric feature parameters and internal component parameters of the fruits, cannot comprehensively describe the fruit information, simply calculates the mass based on the relationship model between the fruit radius and the mass, does not calculate the actual density of the fruits, and does not consider the influence of geometric features and internal components on the density, resulting in inaccurate calculation of the fruit mass. The fixed relationship model adopted is difficult to adapt to the complex and changeable growth environment and fruit individual differences, and cannot accurately capture the complex relationship between the fruit mass and the influencing factors, reducing the accuracy and reliability of yield detection.

[0005] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure. Therefore, it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to provide an oil-tea camellia fruit yield prediction system, method and storage medium based on a three-dimensional laser scanner to solve the problems raised in the above background art.

[0007] To achieve the above purpose, the present invention provides the following technical solutions:

[0008] An oil-tea camellia fruit yield prediction method based on a three-dimensional laser scanner, comprising:

[0009] A data acquisition module, configured to obtain three-dimensional point cloud data of samples and oil-tea camellia fruits to be predicted through a three-dimensional laser scanner, preprocess the three-dimensional point cloud data, perform threshold segmentation on the preprocessed three-dimensional point cloud data to obtain a fruit point cloud region and a branch and leaf point cloud region, obtain the volume of each fruit and geometric feature data affecting fruit density from the fruit point cloud region, collect fruit spectral images of samples and oil-tea camellia fruits to be predicted through a hyperspectral camera, extract fruit spectral data based on the fruit spectral images, select fruits corresponding to the spectral image collection from the collected samples, and use a chemical analysis method to measure the internal components of the fruits to obtain the internal component content data of each fruit;

[0010] A content prediction model construction module, configured to construct a content prediction model based on a deep learning network, use the fruit spectral data of the samples as input and the internal component content data of the fruits as output labels to train the content prediction model;

[0011] A content simulation module, configured to input the fruit spectral data of the oil-tea camellia fruits to be predicted into the trained content prediction model to obtain the internal component content data of the fruits of the oil-tea camellia fruits to be predicted;

[0012] A reference density calculation module, configured to screen multiple standard fruits from the samples as reference samples, calculate the average value of the fruit density of the reference samples according to the density formula, use the average value of the fruit density as the reference density of the fruits, calculate the average values of the geometric feature data and internal component content data of the reference samples, and use the average values of the geometric feature data and internal component content data as the reference values of the geometric feature data and internal component content data;

[0013] A first calculation module, configured to process the geometric feature data and internal component content data of the fruits to generate a density correction index of the fruits;

[0014] A quality prediction model construction module, which is used to construct a fruit quality prediction model based on a deep learning network, taking the reference density, the density correction index of the sample fruits, and the volume as inputs, and the actual quality of each corresponding fruit as the label output, and training the fruit quality prediction model;

[0015] A quality simulation module, which is used to input the reference density, the density correction index of the fruits of the oil-tea camellia tree to be predicted, and the volume into the trained fruit quality prediction model to obtain the actual quality of each fruit of the oil-tea camellia tree to be predicted;

[0016] A second calculation module, which is used to perform an accumulative calculation on the actual quality of each fruit on the oil-tea camellia tree to be predicted to obtain the yield of the oil-tea camellia tree.

[0017] Furthermore, the geometric feature data includes the surface area, diameter, and sphericity of the fruit, and the internal component content data includes the oil content, sugar content, and water content of the fruit.

[0018] Furthermore, the three-dimensional point cloud data of the oil-tea camellia tree is obtained through a three-dimensional laser scanner, and the three-dimensional point cloud data is preprocessed. The preprocessed three-dimensional point cloud data is subjected to threshold segmentation to obtain the fruit point cloud region and the branch and leaf point cloud region. The specific process is as follows:

[0019] Around the oil-tea camellia tree, multiple scanning stations are set at different positions and angles. The stations are evenly distributed around the fruit tree, and the three-dimensional point cloud data is preprocessed, including noise removal and normalization processing;

[0020] The three-dimensional space where the point cloud is located is divided into uniform voxels. The side length of the voxel is determined according to the point cloud density and the segmentation accuracy. The number of point clouds in each voxel is counted, and the mean and standard deviation of the number of point clouds in all voxels are calculated. When the number of point clouds in a certain voxel is greater than the threshold, the threshold is the sum of the mean and the standard deviation, and the point clouds in this voxel belong to the fruit point cloud region. When the number of point clouds in a certain voxel is less than the threshold, the threshold is the difference between the mean and the standard deviation, and it is determined as the branch and leaf point cloud region.

[0021] Furthermore, the geometric feature data and the internal component content data of the fruit are processed to generate the density correction index of the fruit. The formula is as follows:

[0022] ;

[0023] Wherein, is the density correction index of the fruit;

[0024] In the formula, is the surface area of the fruit, is the reference value of the surface area, is the diameter of the fruit, is the reference value of the diameter, is the sphericity of the fruit, is the reference value of sphericity, is the oil content of the fruit, is the reference value of oil content, is the sugar content of the fruit, is the reference value of sugar content, is the water content of the fruit, is the reference value of water content;

[0025] In the formula, is the weight coefficient of the ratio of the surface area to the reference surface area value, is the weight coefficient of the ratio of the diameter to the reference diameter value, is the weight coefficient of the ratio of the sphericity to the reference sphericity value, is the weight coefficient of the ratio of the oil content to the reference oil content value, is the weight coefficient of the ratio of the sugar content to the reference sugar content value, is the weight coefficient of the ratio of the water content to the reference water content value. On the basis of let .

[0026] Furthermore, the actual mass of each fruit on the oil-tea camellia tree to be predicted is accumulated and calculated to obtain the yield of the oil-tea camellia tree. The formula is as follows:

[0027] ;

[0028] Among them, is the yield of the oil-tea camellia tree, is the actual mass of the th fruit, is the index of the fruit on the oil-tea camellia tree to be predicted, , is the number of fruits on the oil-tea camellia tree to be predicted.

[0029] To achieve the above object, the present invention also provides the following technical solution:

[0030] A method for predicting the yield of an oil-tea camellia tree based on a three-dimensional laser scanner. The method is generated based on any one of the above-mentioned systems for predicting the yield of an oil-tea camellia tree based on a three-dimensional laser scanner. The specific steps include:

[0031] S1. Obtain the three-dimensional point cloud data of the sample and the oil-tea camellia fruit trees to be predicted through a three-dimensional laser scanner, preprocess the three-dimensional point cloud data, perform threshold segmentation on the preprocessed three-dimensional point cloud data to obtain the fruit point cloud region and the branch and leaf point cloud region, obtain the volume of each fruit and the geometric feature data affecting the fruit density from the fruit point cloud region, select the volumes and corresponding masses of multiple fruits, calculate the average value of the fruit density, use the average value of the fruit density as the reference density of the fruit, and collect the fruit spectral images of the sample and the oil-tea camellia fruit trees to be predicted through a hyperspectral camera. Based on the fruit spectral images, extract the fruit spectral data. From the collected samples, select the fruits corresponding to the spectral image acquisition, and use chemical analysis methods to determine the internal components of the fruits to obtain the internal component content data of each fruit;

[0032] S2. Build a content prediction model based on a deep learning network, use the fruit spectral data of the sample as the input and the internal component content data of the fruit as the output label to train the content prediction model;

[0033] S3. Input the fruit spectral data of the oil-tea camellia fruit trees to be predicted into the trained content prediction model to obtain the internal component content data of the fruits of the oil-tea camellia fruit trees to be predicted;

[0034] S4. Screen multiple standard fruits from the samples as reference samples, calculate the average value of the fruit density of the reference samples according to the density formula, use the average value of the fruit density as the reference density of the fruit, calculate the average values of the geometric feature data and the internal component content data of the reference samples, and use the average values of the geometric feature data and the internal component content data as the reference values of the geometric feature data and the internal component content data;

[0035] S5. Process the geometric feature data and the internal component content data to generate a density correction index;

[0036] S6. Build a fruit mass prediction model based on a deep learning network, use the reference density, density correction index and volume of the sample fruits as the input and the actual mass of each corresponding fruit as the label output to train the fruit mass prediction model;

[0037] S7. Input the reference density, density correction index and volume of the fruits of the oil-tea camellia fruit trees to be predicted into the trained fruit mass prediction model to obtain the actual mass of each fruit of the oil-tea camellia fruit trees to be predicted;

[0038] S8. Accumulate and calculate the actual mass of each fruit on the oil-tea camellia fruit trees to be predicted to obtain the yield of the oil-tea camellia fruit trees.

[0039] A storage medium for storing a computer program, which when executed by a processor implements any one of the above-mentioned oil-tea camellia tree yield prediction systems based on a three-dimensional laser scanner.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] In the data acquisition stage of the present invention, with the aid of a three-dimensional laser scanner and a hyperspectral camera, data on fruit volume, geometric feature data, and internal component content data are comprehensively obtained. The acquisition of multi-dimensional information can more comprehensively and accurately depict fruit characteristics, laying a foundation for subsequent analysis. A content prediction model is constructed based on a deep learning network to obtain the internal component content data of the oil-tea camellia tree to be predicted. By calculating the density correction index of the fruit and fully considering the influence of various factors on fruit density, the calculation accuracy of fruit quality is effectively improved. A quality prediction model is constructed based on a deep learning network, which can accurately capture the complex relationship between fruit quality and various factors, greatly enhancing the adaptability to different growth environments and fruit individuals. Compared with the simple fixed relationship model in the prior art, this model can better adapt to the complex and changeable growth environment and fruit individual differences, greatly improving the generalization ability of the model. Through the collaborative work of multiple modules, a complete and systematic yield prediction system is formed. From data acquisition, calculation and analysis to model construction and prediction, each link is closely connected. This optimized workflow reduces manual intervention and data errors, and greatly improves the work efficiency of oil-tea camellia tree yield detection. Description of the Drawings

[0042] Figure 1 It is a block diagram of the module composition of the present invention;

[0043] Figure 2 It is a schematic diagram of the overall method flow of the present invention. Detailed Embodiments

[0044] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to specific embodiments.

[0045] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the ordinary meanings understood by those with ordinary skills in the field to which the present invention pertains. The "first", "second" and similar terms used in the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. The terms such as "comprising" or "including" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. The terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left" and "right" are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0046] Embodiment 1:

[0047] Please refer to Figure 1 , the present invention provides a technical solution:

[0048] An oil-tea camellia fruit yield prediction system based on a three-dimensional laser scanner, comprising:

[0049] A data acquisition module, configured to obtain three-dimensional point cloud data of samples and oil-tea camellia fruits to be predicted through a three-dimensional laser scanner, preprocess the three-dimensional point cloud data, perform threshold segmentation on the preprocessed three-dimensional point cloud data to obtain a fruit point cloud region and a branch and leaf point cloud region, obtain the volume of each fruit and geometric feature data affecting fruit density from the fruit point cloud region, collect fruit spectral images of samples and oil-tea camellia fruits to be predicted through a hyperspectral camera, extract fruit spectral data based on the fruit spectral images, select fruits corresponding to the spectral image collection from the collected samples, and use a chemical analysis method to measure the internal components of the fruits to obtain internal component content data of each fruit;

[0050] Based on the above embodiment, the geometric feature data includes the surface area, diameter and sphericity of the fruit, and the internal component content data includes the oil content, sugar content and water content of the fruit.

[0051] Based on the above embodiment, the collected oil-tea camellia fruits are oil-tea camellia fruits in the mature period, so the three-dimensional point cloud image and the fruit spectral image are both images of oil-tea camellia fruits in the mature period.

[0052] Based on the above embodiment, obtaining the three-dimensional point cloud data of the oil-tea camellia fruit through a three-dimensional laser scanner, preprocessing the three-dimensional point cloud data, and performing threshold segmentation on the preprocessed three-dimensional point cloud data to obtain a fruit point cloud region and a branch and leaf point cloud region, the specific process is as follows:

[0053] Set up multiple scanning stations at different positions and angles around the oil-tea camellia tree. The stations are evenly distributed around the tree, and preprocess the three-dimensional point cloud data, including noise removal and normalization processing;

[0054] Divide the three-dimensional space where the point cloud is located into uniform voxels. The side length of the voxel is determined according to the point cloud density and segmentation accuracy. When the point cloud density is high, it is 0.05 meters, and when it is low, it is 0.1 meters. Count the number of point clouds in each voxel, calculate the mean and standard deviation of the number of point clouds in all voxels. When the number of point clouds in a certain voxel is greater than the threshold, which is the sum of the mean and the standard deviation, the point clouds in this voxel are likely to belong to the fruit point cloud area. When the number of point clouds in a certain voxel is less than the threshold, which is the difference between the mean and the standard deviation, it is inclined to be judged as the foliage point cloud area.

[0055] On the basis of the above embodiments, the method for obtaining the volume, surface area, diameter, and sphericity of each fruit from the fruit point cloud area is as follows:

[0056] First, perform triangulation on the fruit point cloud to construct a three-dimensional mesh composed of multiple triangular patches. Then, according to the geometric characteristics of the mesh, use the integral formula to calculate the volume. Assume that the mesh model is composed of triangular patches, the area of the th triangular patch is , is the index of the triangular patch, , the normal vector of the th triangular patch is , the distance from the th triangular patch to the coordinate origin is , then the volume of each fruit is:

[0057] ;

[0058] Convert the fruit point cloud into a triangular mesh through the triangulation algorithm. The mesh is composed of triangular patches. Calculate the area of the th triangular patch. The area of the th triangle is calculated using Heron's formula. Assume that the lengths of the three sides of the th triangle are , , , the semi-perimeter , then the area of the th triangular patch, add the areas of triangular patches to get the surface area ;

[0059] Traverse each pair of points in the fruit point cloud, calculate the Euclidean distance between them. If the coordinates of the two points are and respectively; the Euclidean distance is . Record the maximum distance value as the diameter .

[0060] Calculate the sphericity according to the formula . is the area of the triangular facet, is the volume of each fruit. The closer the value of the sphericity is to 1, the closer the fruit is to a sphere.

[0061] Based on the above embodiments, the specific process of obtaining the number of fruits of the oil tea fruit tree is as follows:

[0062] Use the connected component labeling algorithm to label the voxels in the region determined to be the fruit point cloud. This algorithm starts from an unlabeled fruit voxel, labels it with a specific connected component number, and then recursively labels all other fruit voxels connected to it until all voxels in the connected component are labeled. Then, find the next unlabeled fruit voxel and repeat the above process until all fruit voxels are labeled. Each connected component with a different number represents a potential fruit.

[0063] Calculate the number of voxels contained in each connected component, convert it to volume, and set a minimum volume threshold and a maximum volume threshold according to the actual size range of the oil tea fruit. Connected components with a volume less than the minimum volume threshold are likely to be noise point clouds or misjudged small regions, while connected components with a volume greater than the maximum volume threshold may be multiple fruits sticking together or other abnormal situations. Remove the connected components that do not meet the volume threshold range.

[0064] Perform morphological analysis on the remaining connected components, calculate the sphericity. Since the oil tea fruit is approximately spherical, connected components with a higher sphericity are more likely to be real fruits. Set a sphericity threshold and exclude the connected components with a sphericity lower than this threshold. The number of remaining connected components is the initially counted number of fruits.

[0065] For some overlapping fruits, they may be labeled as one connected component. By analyzing the shape, volume, and internal point cloud distribution characteristics of the connected component, use machine learning algorithms to determine whether the connected component is one fruit or multiple fruits. For example, if the volume of the connected component is significantly larger than the normal volume of the fruit and the internal point cloud distribution shows multiple aggregation centers, it may contain multiple fruits. Split it into multiple fruits for counting according to these characteristics.

[0066] Based on the above embodiments, the spectral images of the fruits of the oil-tea camellia trees are collected by a hyperspectral camera, and the oil content, sugar content, and moisture content of the fruits are extracted from the fruit spectral images. The specific process is as follows:

[0067] For the fruit spectral image, the spectral reflection intensity value of the pixel points in the image is , where represents the horizontal and vertical coordinates of the pixel point. The image is segmented into two parts: the foreground (fruit) and the background. The binary image after segmentation

[0068] is obtained through the following formula:

[0069] When the spectral reflection intensity value of the pixel point is greater than or equal to the spectral reflection intensity value threshold , at this time, the corresponding pixel point is considered as the foreground image, that is, the fruit image;

[0070] When the spectral reflection intensity value of the pixel point is less than the spectral reflection intensity value threshold , the corresponding pixel point is considered as the background image;

[0071] For the fruit region segmented by image processing, the reflectance or absorbance data of this region at each wavelength are extracted, and these data are arranged in the order of wavelength, so as to obtain the spectral curve of this fruit;

[0072] The Soxhlet extraction method is adopted. This method is based on specific chemical principles and experimental operation procedures to process the fruit samples, so as to accurately determine the oil content in the fruits;

[0073] The high performance liquid chromatography method is used. By using the separation and detection capabilities of this method for each component in the mixture, the sugar content in the fruits is measured respectively;

[0074] The drying loss method is used. By drying the fruit samples under certain conditions and measuring the mass change before and after drying, the moisture content in the fruits is determined;

[0075] While measuring the fruit components by chemical methods, the fruits are photographed again with a hyperspectral camera to obtain their spectral images, and then the spectral data corresponding to each fruit are extracted from these spectral images. Since the internal component content of each fruit has been obtained through chemical analysis before, and now there are corresponding spectral data, the internal component content (oil, sugar, moisture) of each fruit and its corresponding spectral information (that is, the reflectance or absorbance data at each wavelength contained in the spectral curve) are successfully corresponded;

[0076] The content prediction model construction module is used to construct a content prediction model based on a deep learning network. Using the fruit spectral data of the sample as the input and the internal component content data of the fruit as the output label, the content prediction model is trained.

[0077] On the basis of the above embodiments, the content prediction model is composed of a deep learning network based on a multi-layer perceptron. The deep neural network of the multi-layer perceptron includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer. The first hidden layer, the second hidden layer, and the third hidden layer all have at least two neurons, and all use ReLU (Rectified Linear Unit) as the activation function.

[0078] In the content prediction model, the input features of the deep learning network of the multi-layer perceptron include: the fruit spectral data of the sample, 1 feature.

[0079] The structure of the deep learning network of the multi-layer perceptron is as follows:

[0080] Input layer: Receives the input of 1 feature.

[0081] First hidden layer: Has 128 neurons and uses ReLU as the activation function.

[0082] Second hidden layer: Has 64 neurons and also uses the ReLU activation function.

[0083] Third hidden layer: Has 32 neurons and uses the ReLU activation function.

[0084] Output layer: Has 1 neuron and outputs the internal component content data of the fruit.

[0085] The process of training the content prediction model is as follows:

[0086] Using the fruit spectral data of the sample as the input quantity and the internal component content data of the fruit as the output label for training, using the mean squared error as the loss function. When the mean squared error is within the range, the training of the content prediction model is completed.

[0087] The content simulation module is used to input the fruit spectral data of the oil tea fruit tree to be predicted into the trained content prediction model to obtain the internal component content data of the fruit of the oil tea fruit tree to be predicted.

[0088] A reference density calculation module is used to screen multiple standard fruits from the samples as reference samples, calculate the average value of the fruit density of the reference samples according to the density formula, take the average value of the fruit density as the reference density of the fruits, calculate the average values of the geometric feature data and the internal component content data of the reference samples, and take the average values of the geometric feature data and the internal component content data as the reference values of the geometric feature data and the internal component content data. The reference values of the geometric feature data include the reference value of the surface area, the reference value of the diameter, and the reference value of the sphericity. The reference values of the internal component content data include the reference value of the oil content, the reference value of the sugar content, and the reference value of the water content.

[0089] The surface area, diameter, sphericity, oil content, sugar content, and water content of the collected fruits need to be normalized. Through normalization, the data of different indicators are unified into the range of 0 to 1, eliminating the influence of dimensions and value ranges, making these data comparable and consistent in subsequent analysis operations.

[0090] A first calculation module is used to process the geometric feature data and the internal component content data to generate a density correction index.

[0091] Based on the above embodiments, the geometric feature data and the internal component content data of the fruits are processed to generate a density correction index for the fruits. The formula is as follows:

[0092] ;

[0093] Among them, is the density correction index of the fruit. The density correction index is used to combine the six indicators of the surface area, diameter, sphericity, oil content, sugar content, and water content of the fruit to comprehensively evaluate the influence of each parameter on the density of each fruit. The larger the density correction index, the greater the influence;

[0094] In the formula, is the surface area of the fruit, is the reference value of the surface area, is the diameter of the fruit, is the reference value of the diameter, is the sphericity of the fruit, is the reference value of the sphericity, is the oil content of the fruit, is the reference value of the oil content, is the sugar content of the fruit, is the reference value of the sugar content, is the water content of the fruit, is the reference value of the water content;

[0095] Select multiple fruits and calculate the average values of surface area, diameter, sphericity, oil content, sugar content, and moisture content. Use the average values of surface area, diameter, sphericity, oil content, sugar content, and moisture content as the reference values for surface area, diameter, sphericity, oil content, sugar content, and moisture content respectively.

[0096] The reference density of the fruit corresponds to the reference values of surface area, diameter, sphericity, oil content, sugar content, and moisture content.

[0097] On this basis, it should be noted that:

[0098] Generally speaking, the higher the sphericity, the closer the fruit is to a spherical shape. For the same volume, the surface area of a sphere is the smallest, the compactness of the internal substances is relatively high, and its density is relatively large when the mass is the same. Therefore, when the sphericity relative to the reference value increases, it indicates that the fruit is closer to a spherical shape, and the density has an increasing trend, which in turn makes the density correction index increase.

[0099] Sugar and moisture are important components inside the fruit. When the sugar content or the moisture content relative to their respective reference values 、 increases, it means that the sugar or moisture inside the fruit increases. When the volume of the fruit changes little, the mass will increase. According to the density formula, the density will increase, which in turn makes the density correction index increase.

[0100] The reference value represents the surface area of the fruit in a certain standard state. When the surface area relative to increases, it means that the shape of the fruit deviates from the standard state and becomes more flattened or irregular. For example, assume that the surface area of a standard spherical fruit is , when it is deformed by an external force, the surface area increases. This change in shape will cause a change in the internal spatial layout of the fruit. When the mass remains unchanged, the volume will relatively increase. From a geometric perspective, for a given volume, a sphere is the shape with the smallest surface area. When the shape of the fruit becomes irregular or flattened, while its surface area increases, in order to accommodate the same mass of substances, the volume will also increase accordingly. According to the density formula, when the mass remains unchanged and the volume increases, the density of the fruit will necessarily decrease. Since the density correction index is used to measure the change in the density of the fruit relative to the standard state, when the surface area relative to When it increases, by affecting the shape and volume of the fruit, it further reduces the density, ultimately resulting in a decrease in the density correction index.

[0101] Base value Represents the standard diameter size of the fruit. When the diameter Relatively Increases, it indicates that the overall size of the fruit has become larger compared to the standard state. For example, the diameter of a certain fruit at a normal growth stage is , if it grows in a special environment and the diameter becomes and is greater than , as the fruit diameter increases, the growth and distribution of its internal cells may change, resulting in a relatively loose internal structure. The cells inside the fruit may be stretched in the diameter direction, and the gaps between cells increase, thus forming more voids inside the fruit. In this case, if the increase in fruit mass is less than the increase in volume caused by the increase in diameter, according to the density formula, the density of the fruit will decrease. Since the density correction index is a quantitative indicator of the fruit density relative to the standard state, when the diameter Relatively Increases, it causes changes in the internal structure of the fruit and a decrease in density, thus causing the density correction index to also decrease accordingly.

[0102] Base value Represents the normal level of the oil content of the fruit. When the oil content Relatively Increases, it indicates that the oil content in the fruit exceeds the normal state. For example, for the fruit of a certain oil crop, the oil content under normal growth conditions is , but after special cultivation or environmental influence, the oil content becomes and is greater than , the oil in the fruit is mainly distributed in the cell gaps or specific oil bodies. As the oil content increases, the oil will expand the cell gaps, increasing the voids inside the fruit, or change the microscopic structure inside the fruit, making it less compact. Since the density of oil is usually less than other components of the fruit, when the fruit mass changes little, the increase in oil content leads to an increase in volume. According to the density formula, the density of the fruit will decrease, and the density correction index is an indicator reflecting the fruit density relative to the standard state. When the oil content Relatively Increases, by changing the internal structure and volume of the fruit, it reduces the density, and thus leads to a decrease in the density correction index.

[0103] In summary, the density correction index is negatively correlated with the surface area, diameter, and oil content, and the density correction index is positively correlated with the sphericity, sugar content, and water content.

[0104] The relative value is obtained by dividing the sugar content of the fruit by the reference value, which can reflect the degree of deviation of the index from the reference state. The reference value indicates the size of the fruit parameter under the set standard state, and is used to measure the size of other fruit parameters relative to the standard. Is the reference value 2 times, , the influence of oil content on the fruit density correction index can be significantly reflected in the formula calculation. Compared with simple numerical values, relative values ​​can better reflect the difference between the actual situation and the standard situation of the indicator. The original numerical values ​​of various indicators of fruits of different types and different growth environments are very different. The benchmark value can be selected according to the research object or application scenario. The formula constructed by dividing by the benchmark value can adapt to different fruit samples. Regardless of the size of the fruit or the content of its ingredients, it can be included in the formula calculation through relative values, making the formula more widely applicable and universal.

[0105] In summary, the above weighted summation formula is used to characterize the density correction index and surface area ,diameter , Sphericity , oil content , sugar content , moisture content The functional relationship between them.

[0106] In the formula, is the weight coefficient of the ratio of surface area to the reference value of surface area, is the weight coefficient of the ratio of diameter to the reference diameter value, is the weight coefficient of the ratio of sphericity to the sphericity reference value, is the weight coefficient of the ratio of fat content to the reference value of fat content, is the weight coefficient of the ratio of sugar content to the reference value of sugar content, is the weight coefficient of the ratio of moisture content to the reference value of moisture content;

[0107] From the perspective of geometric morphology, sphericity fundamentally affects the arrangement and compactness of the internal material of the fruit. The higher the sphericity, the more efficient the internal space utilization of the fruit, the more compact the material distribution, and has a fundamental and structural impact on the density of the fruit. For example, in fruits with high sphericity, components such as cells can be arranged more orderly, reducing internal gaps and directly increasing density.

[0108] The sugar content mainly affects the density from the perspective of component quality. Although the increase in sugar content will increase the fruit quality and thus affect the density, it is only one of the many components of the fruit, and the distribution of sugar in the fruit may be uneven. Its influence on density is more based on the increase in quality, unlike sphericity that affects from the overall structure. Therefore, in terms of the comprehensive influence weight on fruit density, the corresponding to sphericity is greater than the corresponding to sugar content.

[0109] Sugar is a key component in the fruit with a clear mass contribution. Its content change is closely related to fruit maturity, quality, etc., and has a direct and quantifiable impact on fruit density. As the fruit grows, sugar accumulates continuously, and the fruit density will change significantly accordingly.

[0110] The surface area mainly reflects the external characteristics of the fruit. It has no direct causal relationship with the actual content and compactness of the internal substances of the fruit. Even if the surface area changes, the components and structure inside the fruit do not necessarily change accordingly, and its influence on density is relatively indirect. Therefore, the weight corresponding to sugar content is higher than the corresponding to surface area.

[0111] The change in surface area can reflect the wrinkles, concavities and convexities on the fruit surface to a certain extent. These characteristics may affect the interaction between the internal substances of the fruit and the external environment, and then have some influence on the distribution and accumulation of internal substances. For example, a large surface area may be conducive to the accumulation of photosynthesis products and indirectly affect the internal components and density of the fruit.

[0112] The diameter is only a linear measurement of the fruit size. It focuses more on describing the external dimension of the fruit, and its indication of the specific composition and distribution of the internal substances of the fruit is very limited. The increase in diameter may only be the overall expansion of the fruit, which does not mean that the density or composition of the internal substances has a substantial change. Therefore, the weight corresponding to surface area is higher than the corresponding to diameter.

[0113] As a basic measurement of fruit size, the diameter is related to the growth stage and overall development of the fruit to a certain extent. The change trend of fruit diameter can reflect its growth law, which has certain reference value for judging fruit maturity, etc., and maturity is potentially related to fruit density.

[0114] Oils and fats usually exist in the form of small oil droplets dispersed in the fruit. Although the oil and fat content will affect the fruit density, the proportion it accounts for in the fruit components is relatively unstable, and the distribution of oils and fats is relatively random. In contrast, the more macroscopic and stable morphological index of diameter reflects the overall characteristics of the fruit more directly. Therefore, the weight corresponding to diameter corresponding to the oil content being higher 。

[0115] Oil is a relatively stable component in fruits. Although the change in its content has limited impact on fruit density, it has certain regularity and predictability, and oil plays a specific role in the physiological processes and quality characteristics of fruits.

[0116] The water content in fruits is easily affected strongly by external environmental factors (such as air humidity, irrigation conditions, etc.), with a large fluctuation range and relatively random changes. The short-term change in water content may not accurately reflect the physiological state and density characteristics of the fruits themselves, and its impact on fruit density is more of a temporary and unstable factor. Therefore, the weight corresponding to the oil content is higher than that corresponding to the water content 。

[0117] In summary, on the basis of let 。

[0118] As an implementation method, the value range of is 0.1 - 0.3, the value range of is 0.05 - 0.2, the value range of is 0.3 - 0.6,

[0119] The quality prediction model construction module is used to construct a fruit quality prediction model based on a deep learning network, taking the reference density, the density correction index of the sample fruits, and the volume as inputs, and the actual quality of each corresponding fruit as the label output, and training the fruit quality prediction model;

[0120] On the basis of the above embodiments, the quality prediction model is composed of a deep learning network based on a multi-layer perceptron. The deep neural network of the multi-layer perceptron includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer. The first hidden layer, the second hidden layer, and the third hidden layer all have at least two neurons, and all use ReLU as the activation function;

[0121] In the quality prediction model, the input features of the deep learning network of the multi-layer perceptron include: reference density, density correction index, and volume, 3 features.

[0122] The structure of the deep learning network of the multi-layer perceptron is:

[0123] Input layer: Receives inputs of 3 features;

[0124] First hidden layer: Has 128 neurons and uses ReLU as the activation function;

[0125] Second hidden layer: Has 64 neurons and also uses the ReLU activation function;

[0126] Third hidden layer: Has 32 neurons and uses the ReLU activation function;

[0127] Output layer: Has 1 neuron and outputs the actual mass of each fruit.

[0128] The process of training the mass prediction model is as follows:

[0129] Using the reference density, the density correction index of the sample, and the volume as input quantities, and the actual mass of each corresponding fruit as the label for training, and using the mean squared error as the loss function. When the mean squared error is within the range, the training of the mass prediction model is completed.

[0130] Mass simulation module, used to input the reference density, the density correction index of the fruit of the oil-tea camellia tree to be predicted, and the volume into the trained fruit mass prediction model to obtain the actual mass of each fruit of the oil-tea camellia tree to be predicted; Second calculation module, used to perform cumulative calculation on the actual mass of each fruit on the oil-tea camellia tree to be predicted to obtain the yield of the oil-tea camellia tree.

[0131] Based on the above embodiment, by performing cumulative calculation on the actual mass of each fruit on the oil-tea camellia tree to be predicted to obtain the yield of the oil-tea camellia tree, the basis formula is as follows:

[0132] ;

[0133] Wherein, is the yield of the oil-tea camellia tree, is the actual mass of the th fruit, is the index of the fruit on the oil-tea camellia tree to be predicted, , is the number of fruits on the oil-tea camellia tree to be predicted.

[0134] Please refer to Figure 2 , the present invention also provides a technical solution:

[0135] An oil-tea camellia tree yield prediction method based on a 3D laser scanner, the method is generated based on any one of the above-mentioned oil-tea camellia tree yield prediction systems based on a 3D laser scanner, and the specific steps include:

[0136] S1. Obtain the three-dimensional point cloud data of the sample and the oil-tea camellia fruit trees to be predicted through a three-dimensional laser scanner, preprocess the three-dimensional point cloud data, perform threshold segmentation on the preprocessed three-dimensional point cloud data to obtain the fruit point cloud region and the branch and leaf point cloud region, obtain the volume of each fruit and the geometric feature data affecting the fruit density from the fruit point cloud region, and collect the fruit spectral images of the sample and the oil-tea camellia fruit trees to be predicted through a hyperspectral camera. Based on the fruit spectral images, extract the fruit spectral data. From the collected samples, select the fruits corresponding to the spectral image acquisition, and use chemical analysis methods to measure the internal components of the fruits to obtain the internal component content data of each fruit;

[0137] S2. Build a content prediction model based on a deep learning network, use the fruit spectral data of the sample as the input and the internal component content data as the output label to train the content prediction model;

[0138] S3. Input the fruit spectral data of the oil-tea camellia fruit trees to be predicted into the trained content prediction model to obtain the internal component content data of the oil-tea camellia fruit trees to be predicted;

[0139] S4. Screen multiple standard fruits from the samples as reference samples, calculate the average value of the fruit density of the reference samples according to the density formula, use the average value of the fruit density as the reference density of the fruits, calculate the average values of the geometric feature data and the internal component content data of the reference samples, and use the average values of the geometric feature data and the internal component content data as the reference values of the geometric feature data and the internal component content data;

[0140] S5. Process the geometric feature data and the internal component content data to generate a density correction index;

[0141] S6. Build a fruit mass prediction model based on a deep learning network, use the reference density, density correction index and volume of the sample as the input, and the actual mass of each corresponding fruit as the label output to train the fruit mass prediction model;

[0142] S7. Input the reference density, density correction index and volume of the oil-tea camellia fruit trees to be predicted into the trained fruit mass prediction model to obtain the actual mass of each fruit of the oil-tea camellia fruit trees to be predicted;

[0143] S8. Accumulate and calculate the actual mass of each fruit on the oil-tea camellia fruit trees to be predicted to obtain the yield of the oil-tea camellia fruit trees.

[0144] A storage medium for storing a computer program, where the computer program, when executed by a processor, implements any one of the above-mentioned oil-tea camellia fruit tree yield prediction systems based on a three-dimensional laser scanner.

[0145] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data and performing software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0146] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by, electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0147] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, and may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0148] As described above, this is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all of them should be covered by the protection scope of the present application.

Claims

1. A tea tree yield prediction system based on a three-dimensional laser scanner, characterized in that: include: The data acquisition module is used to obtain the three-dimensional point cloud data of the sample and the oil-tea tree to be predicted through a three-dimensional laser scanner, and pre-process the three-dimensional point cloud data, perform threshold segmentation on the pre-processed three-dimensional point cloud data, obtain the fruit point cloud area and the branch and leaf point cloud area, obtain the volume of each fruit and the geometric feature data affecting the fruit density from the fruit point cloud area, and collect the fruit spectral images of the sample and the oil-tea tree to be predicted through a hyperspectral camera, extract the fruit spectral data based on the fruit spectral image, select the fruit corresponding to the spectral image acquisition from the collected samples, use the chemical analysis method to determine the internal components of the fruit, and obtain the internal component content data of each fruit; A content prediction model building module is used to build a content prediction model based on a deep learning network, taking the fruit spectrum data of the sample as input and the internal component content data of the fruit as output labels to train the content prediction model; The content simulation module is used to input the fruit spectral data of the oil-tea tree to be predicted into the trained content prediction model to obtain the internal component content data of the fruit of the oil-tea tree to be predicted; A reference density calculation module is used to select a plurality of standard fruits from the sample as reference samples, calculate the average value of the fruit density of the reference samples according to the density formula, take the average value of the fruit density as the reference density of the fruit, calculate the average value of the geometric characteristic data and the average value of the internal component content data of the reference samples, and take the average value of the geometric characteristic data and the average value of the internal component content data as the reference value of the geometric characteristic data and the internal component content data; The first calculation module is used to process the geometric characteristic data and the internal component content data of the fruit to generate a density correction index of the fruit; The quality prediction model building module is used to build a fruit quality prediction model based on a deep learning network, taking the reference density, density correction index and volume of the sample fruit as input and the corresponding actual mass as the label output to train the fruit quality prediction model; The quality simulation module is used to input the reference density, the density correction index and the volume of the fruit of the oil-tea tree to be predicted into the trained fruit quality prediction model to obtain the actual quality of each fruit of the oil-tea tree to be predicted; The second calculation module is used to accumulate and calculate the actual mass of each fruit on the oil-tea tree to be predicted, so as to obtain the yield of the oil-tea tree.

2. The oil-tea tree yield prediction system based on three-dimensional laser scanner according to claim 1 is characterized in that: The geometric characteristic data include the surface area, diameter and sphericity of the fruit, and the internal component content data include the oil content, sugar content and water content of the fruit.

3. The oil-tea tree yield prediction system based on three-dimensional laser scanner according to claim 1 is characterized in that: The 3D point cloud data of the oil tea tree is obtained by a 3D laser scanner, and the 3D point cloud data is preprocessed. The preprocessed 3D point cloud data is threshold segmented to obtain the fruit point cloud area and the branch and leaf point cloud area. The specific process is as follows: Around the tea trees, multiple scanning stations were set up at different positions and angles. The stations were evenly distributed around the trees, and the 3D point cloud data was pre-processed, including noise removal and normalization. The three-dimensional space where the point cloud is located is divided into uniform voxels. The side length of the voxel is determined according to the point cloud density and segmentation accuracy. The number of point clouds in each voxel is counted, and the mean and standard deviation of the number of point clouds in all voxels are calculated. When the number of point clouds in a voxel is greater than the threshold, the threshold is the sum of the mean and the standard deviation, and the point cloud in the voxel belongs to the fruit point cloud area. When the number of point clouds in a voxel is less than the threshold, the threshold is the difference between the mean and the standard deviation, and it is judged to be the branch and leaf point cloud area.

4. The oil-tea tree yield prediction system based on three-dimensional laser scanner according to claim 2 is characterized in that: The geometric characteristic data of the fruit and the content data of the components are processed to generate the density correction index, based on the following formula: in, is the density correction index of the fruit; In the formula, is the surface area of ​​the fruit, is the reference value of the surface area, is the diameter of the fruit, is the reference value of the diameter, is the sphericity of the fruit, is the reference value of sphericity, is the oil content of the fruit, is the base value of fat content. is the sugar content of the fruit, is the reference value of sugar content. is the moisture content of the fruit, is the base value of moisture content; In the formula, is the weight coefficient of the ratio of surface area to the reference value of surface area, is the weight coefficient of the ratio of diameter to the reference diameter value, is the weight coefficient of the ratio of sphericity to the sphericity reference value, is the weight coefficient of the ratio of fat content to the reference value of fat content, is the weight coefficient of the ratio of sugar content to the reference value of sugar content, is the weight coefficient of the ratio of moisture content to the reference value of moisture content. On the basis of .

5. The oil-tea tree yield prediction system based on three-dimensional laser scanner according to claim 1 is characterized in that: The actual weight of each fruit on the oil-tea tree to be predicted is accumulated and calculated to obtain the yield of the oil-tea tree, based on the following formula: in, The yield of tea oil trees, For the The actual mass of the fruit, is the index of the fruit on the oil-tea tree to be predicted, , The number of fruits of the tea oil tree to be predicted.

6. A method for predicting the yield of oil-tea camellia fruit trees based on a three-dimensional laser scanner, the method being generated based on a system for predicting the yield of oil-tea camellia fruit trees based on a three-dimensional laser scanner as claimed in any one of claims 1 to 5, characterized in that: The specific steps include: S1. Acquire the three-dimensional point cloud data of the sample and the oil-tea tree to be predicted by a three-dimensional laser scanner, pre-process the three-dimensional point cloud data, perform threshold segmentation on the pre-processed three-dimensional point cloud data, obtain the fruit point cloud area and the branch and leaf point cloud area, obtain the volume of each fruit and the geometric feature data affecting the fruit density from the fruit point cloud area, and collect the fruit spectral images of the sample and the oil-tea tree to be predicted by a hyperspectral camera, extract the fruit spectral data based on the fruit spectral image, select the fruit corresponding to the spectral image acquisition from the collected samples, use the chemical analysis method to determine the internal components of the fruit, and obtain the internal component content data of each fruit; S2. Constructing a content prediction model based on a deep learning network, taking the fruit spectral data of the sample as input and the internal component content data of the fruit as output labels, and training the content prediction model; S3. The fruit spectral data of the oil-tea tree to be predicted is input into the trained content prediction model to obtain the internal component content data of the fruit of the oil-tea tree to be predicted; S4. Select multiple standard fruits from the sample as the reference sample, calculate the average value of the fruit density of the reference sample according to the density formula, use the average value of the fruit density as the reference density of the fruit, calculate the average value of the geometric characteristic data and the average value of the internal component content data of the reference sample, and use the average value of the geometric characteristic data and the average value of the internal component content data as the reference value of the geometric characteristic data and the internal component content data; S5. Processing the geometric characteristic data and the internal component content data to generate a density correction index; S6. construct a fruit quality prediction model based on a deep learning network, take the baseline density, density correction index and volume of the sample fruit as input, and the corresponding actual weight of each fruit as the label output, and train the fruit quality prediction model; S7. The baseline density, density correction index and volume of the oil-tea tree fruit to be predicted are input into the trained fruit quality prediction model to obtain the actual quality of each fruit of the oil-tea tree to be predicted; S8. Accumulate and calculate the actual mass of each fruit on the oil-tea tree to be predicted to obtain the yield of the oil-tea tree.

7. A storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, implements a tea oil tree yield prediction system based on a three-dimensional laser scanner as described in any one of claims 1-5.

Images