Cotton field boll opening rate remote sensing monitoring method and device based on neural network
By using a neural network-based method, combined with drone image acquisition and deep learning, the number of cotton bolls with boll opening was identified and mathematical modeling was performed, which solved the problems of low efficiency and accuracy in cotton field boll opening rate monitoring and achieved efficient and accurate monitoring in complex environments.
Patent Information
- Application Number
- CN202510761939.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-09
- Publication Date
- 2025-10-17
AI Technical Summary
The efficiency and accuracy of cotton field boll opening rate monitoring in existing technologies are low, and accurate monitoring is difficult to achieve, especially in real complex cotton field environments with low defoliation rates and severe obstruction.
A neural network-based method was adopted, combined with UAV image acquisition, structural functional model and deep learning. The number of cotton bolls with boll opening was identified by training the model, and the target relationship function was used for mathematical modeling to determine the boll opening rate of the cotton field.
The accuracy and efficiency of cotton field boll opening rate monitoring are improved, and it can realize fast and accurate calculation of boll opening rate in cotton field environments with low defoliation rate and severe shading, saving labor costs.
Smart Images
Figure CN120808098A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of crop growth monitoring, and particularly relates to a cotton boll shedding rate remote sensing monitoring method and device based on a neural network. BACKGROUND
[0002] The cotton boll shedding rate refers to the ratio of the boll shedding number to the total boll number (the sum of the boll shedding number and the non-boll shedding number) of a single cotton plant, and is used to represent the maturity process in the late growth period of a cotton field.
[0003] In the prior art, the cotton boll shedding rate is mostly monitored based on an unmanned aerial vehicle (UAV) remote sensing image, and the monitoring methods mainly include the following two methods: (1) a vegetation index or a vegetation index change rate is calculated by using a UAV multispectral image, and then the cotton boll shedding rate is monitored, this method has poor migration performance and low monitoring accuracy; (2) an RGB image of a cotton field is obtained by using a UAV, three-dimensional point cloud of the cotton field is obtained, and then a deep learning algorithm is used to segment, classify and count the cotton bolls in the point cloud, and the cotton boll shedding rate is calculated, this method has a complex processing procedure and low monitoring efficiency. SUMMARY
[0004] The present application provides a cotton boll shedding rate remote sensing monitoring method and device based on a neural network, which is used to solve the technical problem of low efficiency and accuracy in monitoring the cotton boll shedding rate in the prior art.
[0005] The present application provides a cotton boll shedding rate remote sensing monitoring method based on a neural network, which comprises the following steps: an image of a current cotton field collected by a UAV is obtained; the current cotton field image is input into a first model, and the number of boll shedding cotton bolls in the current cotton field image output by the first model is obtained; wherein the first model is obtained by training a first historical cotton field image as sample data and corresponding labeled boll shedding cotton boll number as label data; based on the number of boll shedding cotton bolls, the boll shedding cotton boll density in the current cotton field image is determined; based on the boll shedding cotton boll density and a pre-determined target relationship function, the cotton boll shedding rate is determined; the target relationship function is a relationship function between the boll shedding cotton boll density and the cotton boll shedding rate.
[0006] According to the cotton boll shedding rate remote sensing monitoring method based on a neural network provided by the present application, the first historical cotton field image comprises a historical cotton field image collected by a UAV and a cotton field simulation image generated by a second model; the second model is obtained by training a cotton field overhead image generated by a CottonXL model and a historical cotton field image collected by a UAV as training data; wherein the positions of the boll shedding cotton bolls in the cotton field overhead image and the cotton field simulation image are consistent.
[0007] According to the cotton field boll shedding rate remote sensing monitoring method based on the neural network provided by the application, the step of generating the cotton field simulation image by the second model comprises: The cotton field overhead image to be converted is input into the second model, and a cotton field simulation image generated by the second model is obtained.
[0008] According to the cotton field boll shedding rate remote sensing monitoring method based on the neural network provided by the application, the determination of the position of the boll shedding cotton boll in the cotton field overhead image and the cotton field simulation image comprises: The cotton field overhead image is input into a third model, and a cotton field overhead image with a boll shedding cotton boll position label output by the third model is obtained; wherein the third model is obtained by training with the cotton field overhead image as sample data and corresponding labeled boll shedding cotton boll positions as label data; The labeled position of the boll shedding cotton boll in the cotton field overhead image is used as the common position of the boll shedding cotton boll in the cotton field overhead image and the cotton field simulation image.
[0009] According to the cotton field boll shedding rate remote sensing monitoring method based on the neural network provided by the application, the step of generating the cotton field overhead image by the CottonXL model comprises: Obtaining growth information of cotton in a cotton field; the growth information comprises cotton variety, sowing time, planting density, plant row configuration, topping time, number of cotton fruiting branches, and meteorological data of the whole growth period; Parameter setting is performed on the CottonXL model according to the growth information, and a parameter corrected CottonXL model is obtained; the parameter corrected CottonXL model is used to output a cotton field overhead image according to the growth information; The cotton field overhead image meeting the preset condition is output by the parameter corrected CottonXL model; wherein the preset condition comprises the number and type of the cotton field overhead image.
[0010] According to the cotton field boll shedding rate remote sensing monitoring method based on the neural network provided by the application, the determination of the target relationship function comprises: The number of boll shedding cotton bolls in the cotton field overhead image is output by the CottonXL model, and the boll shedding cotton boll density in the cotton field overhead image is calculated; Based on the corresponding relationship between the boll shedding cotton boll density and the cotton field boll shedding rate in the cotton field overhead image output by the CottonXL model, the target relationship function is determined.
[0011] The application further provides a cotton field boll shedding rate remote sensing monitoring device based on a neural network, comprising the following modules: An acquisition module is configured to acquire a current cotton field image collected by a UAV; The monitoring module is configured to input the current cotton field image into the first model to obtain a number of bolls in the current cotton field image output by the first model, wherein the first model is trained by taking a first historical cotton field image as sample data and taking a corresponding labeled number of bolls as label data. The first determining module is configured to determine a boll density in the current cotton field image based on the number of bolls. The second determining module is configured to determine a cotton field boll shedding rate based on the boll density and a predetermined target relationship function, wherein the target relationship function is a relationship function between the boll density and the cotton field boll shedding rate.
[0012] The application further provides an electronic device including a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the neural network-based cotton field boll shedding rate remote sensing monitoring method according to any one of the above when executing the computer program.
[0013] The application further provides a non-transitory computer-readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the neural network-based cotton field boll shedding rate remote sensing monitoring method according to any one of the above.
[0014] The application further provides a computer program product including a computer program, and the computer program is executed by a processor to implement the neural network-based cotton field boll shedding rate remote sensing monitoring method according to any one of the above.
[0015] The application provides a neural network-based cotton field boll shedding rate remote sensing monitoring method and device, which inputs a current cotton field image collected by a UAV into a first model to obtain a number of bolls in the current cotton field image output by the first model, wherein the first model is trained by taking a first historical cotton field image as sample data and taking a corresponding labeled number of bolls as label data, and the generalization performance of the model in a real complex cotton field environment is improved through pre-training, so as to improve the recognition accuracy of the model and further improve the subsequent cotton field boll shedding rate monitoring accuracy; then a boll density in the current cotton field image is determined based on the number of bolls, and a cotton field boll shedding rate is determined based on the boll density and a predetermined target relationship function, so that the mathematical modeling from the boll density to the cotton field boll shedding rate is realized through the target relationship function, the cotton field boll shedding rate is determined without complex statistical calculation, the calculation efficiency is improved, and the cotton field boll shedding rate monitoring efficiency is further improved. BRIEF DESCRIPTION OF DRAWINGS
[0016] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.
[0017] Figure 1 is a flowchart of a cotton boll shedding rate remote sensing monitoring method based on a neural network provided by the present application.
[0018] Figure 2 is a relationship curve diagram between actual boll shedding rate and boll shedding density provided by the present application.
[0019] Figure 3 is a schematic diagram of the CottonXL model generating an overhead view, the CycleGAN model generating a virtual image, and the overhead view actually taken by a drone provided by the present application.
[0020] Figure 4 is a schematic diagram of the CottonXL model simulating an overhead view of a cotton field provided by the present application.
[0021] Figure 5 is a structural schematic diagram of a cotton boll shedding rate remote sensing monitoring device based on a neural network provided by the present application.
[0022] Figure 6 is a structural schematic diagram of an electronic device provided by the present application. DETAILED DESCRIPTION
[0023] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely in combination with the drawings in the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort belong to the protection scope of the present application.
[0024] In recent years, the level of cotton production mechanization has rapidly increased, and the cultivation link has basically achieved mechanization, and the harvesting link has also shown a rapid development trend. The application of harvesting aids for defoliation and maturation is the premise of mechanical harvesting of cotton, and the determination of the spraying time and dosage of harvesting aids needs to consider a variety of factors, among which the shedding rate is a key factor. The shedding rate refers to the ratio of shed cotton bolls to total bolls (the sum of shed bolls and unshed bolls) on a single cotton plant, which is used to represent the maturation process in the late growth period of the cotton field. In addition, the shedding rate is also an important indicator for evaluating the application effect of harvesting aids and judging the mechanical harvesting time. Generally, the cotton field shedding rate should reach more than 90% before mechanical harvesting. As an important indicator in the late growth period of cotton field production, the shedding rate is mostly obtained by traditional investigation methods, i.e., first selecting sample points and determining the investigated plants, then counting the shed bolls and unshed bolls of each plant, and finally calculating the shedding rate by dividing the number of shed bolls by the total number of bolls (the sum of shed bolls and unshed bolls). This method has high labor intensity, long time consumption, low efficiency, and limited number of sample points investigated per unit time, and thus has poor representativeness. Therefore, it is urgent to develop a method for quickly and accurately monitoring the shedding rate of cotton fields.
[0025] Most of the prior art methods for predicting the shedding rate are based on unmanned aerial vehicle remote sensing images, and the main methods include the following two methods: (1) vegetation index inversion, and (2) combination of three-dimensional point cloud and deep learning. Method (1) uses unmanned aerial vehicle multispectral images to calculate the vegetation index or vegetation index change rate, and models the real value of the ground investigation shedding rate to realize the monitoring of the shedding rate. Method (2) uses an unmanned aerial vehicle to obtain multi-angle RGB images of a cotton field by cross-ring flight before harvesting, obtains three-dimensional point cloud of the cotton field based on these images, realizes group reconstruction at the organ level, and then uses a deep learning algorithm to segment, classify and count the cotton bolls in the point cloud to calculate the shedding rate of the cotton field.
[0026] However, in the above method (1), the data set used for modeling usually has insufficient scene richness, and can only maintain high accuracy in a specific scene, and the scene migration ability is poor; in the above method (2), there are problems such as low monitoring efficiency, complex processing flow and short applicable period, for example, this method needs about 2 hours to realize data acquisition of 4500 m 2 , and the subsequent full-process processing also needs 6 hours of processing time, which is difficult to meet the required time in production, so this method is currently only applicable to cotton fields with high defoliation rate and less serious shading.
[0027] In summary, whether it is cotton boll recognition or shedding rate prediction, the prior art is not applicable to real complex cotton field environments with low defoliation rate and serious shading, and has poor scene migration ability, making it difficult to achieve real application.
[0028] To this end, the application provides a cotton field boll shedding rate remote sensing monitoring method based on a neural network, which combines a structure function model and deep learning to realize accurate estimation of the cotton field boll shedding rate in the whole stage of the late growth period of cotton, so as to provide a basis for accurate application of leaf shedding and ripening agents in the late growth period of cotton, evaluation of ripening effect, and selection of harvesting time.
[0029] The application will be described below in combination with Figures 1 to 6 The application provides a cotton field boll shedding rate remote sensing monitoring method and device based on a neural network.
[0030] Figure 1 A flowchart of the cotton field boll shedding rate remote sensing monitoring method based on a neural network provided by the application is shown in FIG. Figure 1 The method comprises the following steps: Step 101: acquiring a current cotton field image collected by a UAV.
[0031] Specifically, in the embodiment of the application, multi-angle RGB images of the cotton field to be monitored are collected by a UAV as subsequent model input images.
[0032] Step 102: inputting the current cotton field image into a first model to obtain the number of bolls shed in the current cotton field image output by the first model; wherein the first model is obtained by training with first historical cotton field images as sample data and corresponding labeled boll shedding numbers as label data.
[0033] Specifically, the first model is pre-trained in a supervised learning manner with first historical cotton field images as sample data and corresponding labeled boll shedding numbers as label data, wherein the first model can be various target detection models in the prior art and trained, such as a trained Yolo series deep learning model. In the embodiment of the application, a trained Yolov11 model is taken as the first model as an example for description.
[0034] Through the above pre-training, the generalization performance of the Yolov11 model in a real complex cotton field environment is improved, so that the Yolov11 model can cope with a real complex cotton field environment with low leaf shedding rate and serious occlusion, the scene migration ability of the Yolov11 model is improved, and thus the model recognition accuracy is improved.
[0035] After the Yolov11 model is trained in the above manner, the current cotton field image collected by the UAV is input into the Yolov11 model to obtain the number of bolls shed in the current cotton field image output by the Yolov11 model, so that the number of bolls shed in the cotton field can be accurately and efficiently monitored in combination with the image collected by the UAV remote sensing, and thus the monitoring accuracy and efficiency of the boll shedding rate of the cotton field are improved.
[0036] Step 103: Determine the density of opened cotton bolls in the current cotton field image based on the number of opened cotton bolls.
[0037] Specifically, after determining the number of open cotton bolls in the current cotton field image collected by the UAV, further calculation is performed based on the ground sampling distance (GSD) of the UAV image to obtain the open cotton boll density in the current cotton field image.
[0038] Where DOB (Density of Opening Bolls) represents the density of open bolls in the current cotton field image, NOB (Numbers of Opening Bolls) represents the number of open bolls in the current cotton field image, A represents the corresponding actual area in the current cotton field image; W represents the number of pixels on the wide side of the current cotton field image, and H represents the number of pixels on the long side of the current cotton field image.
[0039] Step 104: Determine the cotton field boll opening rate based on the boll opening density and a predetermined target relationship function; the target relationship function is a relationship function between the boll opening density and the cotton field boll opening rate.
[0040] Optionally, determining the target relationship function includes: Output the number of opened cotton bolls in the cotton field top view by using the CottonXL model, and calculate the density of opened cotton bolls in the cotton field top view; The target relationship function is determined based on the corresponding relationship between the cotton boll opening density and the cotton field opening rate in the cotton field top view output by the CottonXL model.
[0041] Specifically, a cotton field top view was simulated in advance using a structural functional model (i.e., the CottonXL model). The cotton field top view simulated by the CottonXL model was labeled with the number of opened cotton bolls and the cotton field's opening rate. The open cotton boll density in the cotton field top view was then calculated based on the labeled number of opened cotton bolls and the GSD parameter information set when the CottonXL model was used to simulate the cotton field top view. Finally, data fitting was performed based on the labeled cotton field's opening rate and the calculated open cotton boll density to obtain the nonlinear relationship between the cotton field's opening rate and the open cotton boll density, as well as the corresponding relationship function (i.e., the target relationship function). Figure 2 Schematic diagram of the relationship between the actual cotton boll opening rate and the cotton boll opening density provided by the present invention. Figure 2 shown.
[0042] In the formula, BOR (Boll Opening Rate) represents the actual shedding rate, DOB represents the shedding boll density, and A, B, C and D are empirical coefficients obtained through curve fitting.
[0043] The embodiment of the application obtains a target relationship function between the shedding boll density of the overhead view of the cotton field and the shedding rate of the cotton field (i.e., the actual shedding rate) according to the CottonXL model, realizes mathematical modeling from local image features to the shedding rate of the cotton field, and makes it possible to quickly and accurately calculate the shedding rate of the cotton field by combining the shedding boll density in the current cotton field image collected by the unmanned aerial vehicle during the monitoring process.
[0044] In some embodiments, the image of the entire cotton field is collected by the unmanned aerial vehicle, the shedding boll density of the entire cotton field is obtained, the shedding rate of the entire cotton field and the positions of the cotton field corresponding to different shedding rates are further calculated, and thus the shedding rate distribution map of the entire cotton field is obtained without tedious investigation, sampling, statistical analysis, which greatly improves the efficiency and accuracy of the shedding rate monitoring, saves labor costs, realizes continuous tracking of the shedding dynamics of the cotton in the later growth period, and provides a quantitative basis for precision harvesting decisions.
[0045] The application provides a cotton field shedding rate remote sensing monitoring method based on a neural network, which inputs the current cotton field image collected by the unmanned aerial vehicle into the first model to obtain the number of shedding bolls in the current cotton field image output by the first model, wherein the first model is trained by taking the first historical cotton field image as sample data and taking the corresponding labeled number of shedding bolls as label data, the generalization performance of the model in the real complex cotton field environment is improved through pre-training, the identification accuracy of the model is improved, and thus the subsequent cotton field shedding rate monitoring accuracy is improved; then the shedding boll density in the current cotton field image is determined based on the number of shedding bolls, and the shedding rate of the cotton field is determined based on the shedding boll density and the pre-determined target relationship function, so that the mathematical modeling from the shedding boll density to the shedding rate of the cotton field is realized through the target relationship function, the shedding rate of the cotton field is determined without complex statistical calculation, the calculation efficiency is improved, and thus the cotton field shedding rate monitoring efficiency is improved.
[0046] Optionally, the first historical cotton field image includes a historical cotton field image collected by the unmanned aerial vehicle and a cotton field simulation image generated by the second model. The second model is trained by taking the overhead view of the cotton field generated by the CottonXL model and the historical cotton field image collected by the unmanned aerial vehicle as training data; wherein the positions of the shedding bolls in the overhead view of the cotton field and the cotton field simulation image remain consistent.
[0047] Specifically, before training the Yolov11 model, not only a certain number of historical cotton field images are collected by the unmanned aerial vehicle, but also a large number of cotton field simulation images are generated by the second model to expand the training data. The second model is a kind of deep learning model for generating virtual images and is trained, which can be a trained GAN model, a CycleGAN model, etc. In the embodiment of the present application, a trained CycleGAN model is taken as the second model as an example for illustration.
[0048] The CycleGAN model is supervised trained with the cotton field overhead images generated by the CottonXL model and the historical cotton field images collected by the unmanned aerial vehicle as training data, so that the trained CycleGAN model can convert the simulated cotton field overhead images into cotton field simulation images (i.e. virtual images), and the cotton field simulation images can reach the level of deception, that is, the cotton field simulation images are highly similar to the real cotton field images collected by the unmanned aerial vehicle. In addition, in order to improve the conversion effect, a related loss function (such as mean square error loss function) needs to be set in the conversion process, so that the positions of key elements (such as shedding bolls) in the cotton field overhead images before conversion and the cotton field simulation images after conversion remain consistent, thereby improving the usability of the converted cotton field simulation images.
[0049] Optionally, the determination of the positions of the shedding bolls in the cotton field overhead images and the cotton field simulation images comprises: inputting the cotton field overhead images into a third model to obtain cotton field overhead images with shedding boll position labels output by the third model; wherein the third model is trained with the cotton field overhead images as sample data and corresponding labeled shedding boll positions as label data; taking the labeled positions of the shedding bolls in the cotton field overhead images as the common positions of the shedding bolls in the cotton field overhead images and the cotton field simulation images.
[0050] Specifically, the third model can be any kind of deep learning model for target detection and trained, such as a variety of deep learning models of the Yolo series trained. In the embodiment of the present application, a trained Yolov11 model is taken as the third model as an example for illustration. When the Yolov11 model is supervised trained for position labeling, first, cotton field overhead images are generated by the CottonXL model and taken as sample data, then the positions of the bolls in the cotton field overhead images are manually labeled to obtain label data, and finally the Yolov11 model is supervised trained with the above sample data and label data to obtain the third model.
[0051] The cotton field overhead image is input into the third model to obtain a cotton field overhead image with boll shedding cotton boll position labels output by the third model; and the labeled position of the boll shedding cotton boll in the cotton field overhead image is taken as the common position of the boll shedding cotton boll in the cotton field overhead image and the cotton field simulation image.
[0052] Further, after outputting the cotton field overhead image with accurate coordinates by the third model, the cotton field overhead image with accurate coordinates is input into the trained CycleGAN model, so that the trained CycleGAN model can retain the boll shedding cotton boll position feature in the process of converting the simulated cotton field overhead image into the cotton field simulation image, thereby generating a high-fidelity cotton field simulation image and improving the reliability of the converted cotton field simulation image.
[0053] Optionally, the step of generating the cotton field simulation image by the second model comprises: inputting the cotton field overhead image to be converted into the second model to obtain the cotton field simulation image generated by the second model.
[0054] Specifically, Figure 3 is a schematic diagram of the CottonXL model, the CycleGAN model and the overhead view of the real aerial photograph provided by the present application, as Figure 3 shown, first, the cotton field overhead view simulated by the CottonXL model is taken as the input of the trained CycleGAN model, then the cotton field overhead view is converted in style by the trained CycleGAN model, and finally the cotton field simulation image is generated, which is highly similar to the overhead view of the real aerial photograph, and can achieve a false appearance of truth, as usable training data.
[0055] Based on the above embodiment, when training the first model, the historical cotton field images collected by the unmanned aerial vehicle and the cotton field simulation images generated by the second model can also be mixed according to a predetermined proportion to obtain an expanded data set.
[0056] For example, in the case of 1000 historical cotton field images collected by the unmanned aerial vehicle and cotton field simulation images generated by the trained CycleGAN model, the mixed images of the historical cotton field images collected by the unmanned aerial vehicle and the cotton field simulation images generated by the trained CycleGAN model are used as sample data (i.e. the first historical cotton field image) in 11 proportions of 10:0, 9:1, 8:2, 7:3, 6:4, 5:5, 4:6, 3:7, 2:8, 1:9 and 0:10, and the corresponding labeled boll number is used as label data to train the Yolov11 model. According to the precision data obtained by training, the optimal proportion of the historical cotton field images collected by the unmanned aerial vehicle and the cotton field simulation images generated by the CycleGAN model is determined, and then the Yolov11 model trained at the optimal proportion is used as the first model, so as to obtain the best performance model for subsequent identification and detection of the boll number in the current cotton field image collected by the unmanned aerial vehicle, and further improve the identification accuracy.
[0057] In addition, when labeling the boll number in the first historical cotton field image, since the CottonXL model simulates the cotton field overhead image with labeled information of the boll number, the cotton field simulation image generated by the trained CycleGAN model also has the labeled information of the boll number. Therefore, when training the Yolov11 model, it is not necessary to manually label the cotton field simulation image generated by the trained CycleGAN model, so as to further save the labor cost and improve the efficiency while realizing the expansion of the data set.
[0058] The present application uses the CottonXL model and the unmanned aerial vehicle remote sensing data to design a virtual-real data mixed training mechanism (11 proportion optimization), expands the training data set, solves the problem of lack of labeled data in the agricultural scene, and breaks through the generalization bottleneck of the traditional model in the real complex farmland environment by screening the optimal fusion proportion of the unmanned aerial vehicle image and the simulation image, so as to obtain the best performance model, so that the first model can cope with the real complex cotton field environment with low leaf shedding rate and serious shielding, improve the scene migration ability of the model, and thus improve the model recognition accuracy.
[0059] Optionally, the step of generating the cotton field overhead view by the CottonXL model comprises: obtaining growth information of the cotton in the cotton field; the growth information comprises cotton variety, sowing time, planting density, plant row configuration, topping time, number of fruiting branches, and meteorological data of the whole growth period; performing parameter setting on the CottonXL model according to the growth information to obtain a parameter-corrected CottonXL model; the parameter-corrected CottonXL model is used to output a cotton field overhead view according to the growth information. The CottonXL model after the parameter correction outputs a cotton field overhead view meeting preset conditions; wherein, the preset conditions include the number and category of the cotton field overhead view.
[0060] Specifically, the CottonXL model is an existing agricultural functional structure model specially used for simulating the growth and development process of cotton, and is mainly used for precision management in cotton production. Figure 4 is a schematic diagram of the CottonXL model simulating a cotton field overhead view provided by the present application, as Figure 4 indicated.
[0061] Before using the CottonXL model, it is usually necessary to correct the parameters of the CottonXL model, reduce the prediction error caused by improper parameter setting, improve the accuracy and reliability of the CottonXL model, and ensure that the CottonXL model can stably and efficiently run in various application scenarios.
[0062] In the embodiments of the present application, first, the growth information of the cotton in the cotton field to be monitored needs to be obtained, such as cotton variety, sowing time, planting density, plant row configuration, topping time, number of cotton fruiting branches, and meteorological data of the whole growth period; then, the CottonXL model is set according to the growth information, and the CottonXL model after parameter correction is obtained, so that the model can simulate an accurate cotton field overhead view meeting preset conditions for the cotton field to be monitored.
[0063] In some embodiments, in order to make the cotton field overhead view simulated by the CottonXL model more consistent with the real complex cotton field environment, a plurality of groups of random numbers can also be set to make the cotton field overhead view simulated by the CottonXL model produce random changes within a certain range rather than being completely the same, for example, simulate an overhead view changing with the cotton growth time, an overhead view changing with the difference in cotton variety in different cotton field regions, and the like, so as to obtain a more rich and accurate data set, and further improve the generalization ability and recognition accuracy of the model after subsequent training.
[0064] Next, a cotton field boll opening rate remote sensing monitoring device based on a neural network provided by the present application is described, and the cotton field boll opening rate remote sensing monitoring device based on a neural network described below can be correspondingly referred to the cotton field boll opening rate remote sensing monitoring method based on a neural network described above.
[0065] Based on any one of the above embodiments, Figure 5 is a structural schematic diagram of a cotton field boll opening rate remote sensing monitoring device based on a neural network provided by the present application, as Figure 5As shown. The embodiment of the application provides a cotton field boll shedding rate remote sensing monitoring device based on a neural network, which comprises an acquisition module 501, a monitoring module 502, a first determination module 503 and a second determination module 504, wherein: The acquisition module 501 is used for acquiring a current cotton field image collected by a UAV; the monitoring module 502 is used for inputting the current cotton field image into a first model to obtain a number of bolls shedding in the current cotton field image output by the first model; wherein the first model is obtained by training with a first historical cotton field image as sample data and with a corresponding labeled boll shedding number as label data; the first determination module 503 is used for determining a boll shedding density in the current cotton field image based on the number of bolls shedding; and the second determination module 504 is used for determining a cotton field boll shedding rate based on the boll shedding density and a predetermined target relationship function; the target relationship function is a relationship function between the boll shedding density and the cotton field boll shedding rate.
[0066] The application provides a cotton field boll shedding rate remote sensing monitoring device based on a neural network, which inputs a current cotton field image collected by a UAV into a first model to obtain a number of bolls shedding in the current cotton field image output by the first model, wherein the first model is obtained by training with a first historical cotton field image as sample data and with a corresponding labeled boll shedding number as label data, the generalization performance of the model in a real complex cotton field environment is improved through pre-training, so that the model recognition accuracy is improved, and then the subsequent cotton field boll shedding rate monitoring accuracy is improved; then a boll shedding density in the current cotton field image is determined based on the number of bolls shedding, and a cotton field boll shedding rate is determined based on the boll shedding density and a predetermined target relationship function, so that the mathematical modeling from the boll shedding density to the cotton field boll shedding rate is realized through the target relationship function, the cotton field boll shedding rate can be determined without complex statistical calculation, the calculation efficiency is improved, and then the cotton field boll shedding rate monitoring efficiency is improved.
[0067] Figure 6 An example of an entity structure diagram of an electronic device is shown as Figure 6 The electronic device can include a processor 610, a communications interface 620, a memory 630 and a communications bus 640, wherein the processor 610, the communications interface 620 and the memory 630 complete mutual communication through the communications bus 640. The processor 610 can invoke a logic instruction in the memory 630 to execute a cotton field boll shedding rate remote sensing monitoring method based on a neural network, which comprises: Acquiring a current cotton field image collected by a UAV; input the current cotton field image into the first model to obtain a number of bolls in the current cotton field image output by the first model, wherein the first model is obtained by training with first historical cotton field images as sample data and corresponding labeled numbers of bolls as label data; determine a boll shedding density in the current cotton field image based on the number of bolls; determine a cotton field boll shedding rate based on the boll shedding density and a predetermined target relationship function, wherein the target relationship function is a relationship function between the boll shedding density and the cotton field boll shedding rate.
[0068] In addition, the logical instructions in the memory 630 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.
[0069] On the other hand, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute the neural network-based cotton field boll shedding rate remote sensing monitoring method provided by the above-mentioned method, which comprises: obtaining a current cotton field image collected by a UAV; inputting the current cotton field image into the first model to obtain a number of bolls in the current cotton field image output by the first model, wherein the first model is obtained by training with first historical cotton field images as sample data and corresponding labeled numbers of bolls as label data; determine a boll shedding density in the current cotton field image based on the number of bolls; determine a cotton field boll shedding rate based on the boll shedding density and a predetermined target relationship function, wherein the target relationship function is a relationship function between the boll shedding density and the cotton field boll shedding rate.
[0070] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the neural network-based cotton field boll opening rate remote sensing monitoring method provided by the above methods, the method comprising: Get the current cotton field images collected by drone; Inputting the current cotton field image into a first model, and obtaining the number of opened cotton bolls in the current cotton field image output by the first model; wherein the first model is trained using the first historical cotton field image as sample data and the corresponding annotated number of opened cotton bolls as label data; Determining a density of opened cotton bolls in the current cotton field image based on the number of opened cotton bolls; The cotton field boll opening rate is determined based on the boll opening density and a predetermined target relationship function; the target relationship function is a relationship function between the boll opening density and the cotton field boll opening rate.
[0071] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.
[0072] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.
[0073] It should be noted that, as used in this document, the terms "includes," "including," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements is not required to only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element. Additionally, it should be noted that the scope of the methods and apparatus of the present embodiments are not limited by the order of the steps or the order of the sequence, and can include performing the steps in different order than described, or in substantially simultaneous fashion, or in reverse order, such as described, and can also include adding, omitting, or combining various steps. Additionally, features described in relation to certain examples can be combined in other examples.
[0074] It should also be noted that the terms "target", "first", "second", etc. are used to distinguish similar objects, and are not intended to describe a particular order or sequence. It should be understood that the terms used in this way can be interchanged as appropriate, so that the embodiments of the present application can be implemented in an order other than that illustrated or described herein, and the objects distinguished by "first", "second" are generally of a kind and are not limited in number, for example, the first object can be one or more.
[0075] "Determining B based on A" in the embodiments of the present application means that A is considered as a factor when determining B. It is not limited to "determining B based on A only", but also includes "determining B based on A and C", "determining B based on A, C and E", "determining C based on A, and determining B based on C further", etc. In addition, it can also include A as a condition for determining B, for example, "when A meets the first condition, determining B using the first method"; for example, "when A meets the second condition, determining B"; for example, "when A meets the third condition, determining B based on the first parameter"; etc. Of course, A can also be a condition for determining B, for example, "when A meets the first condition, determining C using the first method, and further determining B based on C", etc.
[0076] The term "a plurality of" in the present application means two or more, and other quantifiers are similar.
[0077] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; and although the present application has been described in detail with reference to the foregoing embodiments, it should be appreciated by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A remote sensing monitoring method for cotton field boll opening rate based on neural network, characterized in that: include: Get the current cotton field images collected by drone; Inputting the current cotton field image into a first model, and obtaining the number of opened cotton bolls in the current cotton field image output by the first model; wherein the first model is trained using the first historical cotton field image as sample data and the corresponding annotated number of opened cotton bolls as label data; Determining a density of opened cotton bolls in the current cotton field image based on the number of opened cotton bolls; The cotton field boll opening rate is determined based on the boll opening density and a predetermined target relationship function; the target relationship function is a relationship function between the boll opening density and the cotton field boll opening rate.
2. The method for remote sensing monitoring of cotton field boll opening rate based on neural network according to claim 1, characterized in that: The first historical cotton field image includes a historical cotton field image collected by a drone and a cotton field simulation image generated by a second model; The second model is trained using a cotton field overhead image generated by the CottonXL model and historical cotton field images collected by drones as training data; wherein the position of the popping cotton bolls in the cotton field overhead image and the cotton field simulation image remains consistent.
3. The method for remote sensing monitoring of cotton field boll opening rate based on neural network according to claim 2, characterized in that: The step of generating a cotton field simulation image by the second model includes: The overhead image of the cotton field to be converted is input into the second model to obtain a simulated image of the cotton field generated by the second model.
4. The method for remote sensing monitoring of cotton field boll opening rate based on neural network according to claim 2, characterized in that: Determining the position of the cotton boll in the cotton field overhead image and the cotton field simulation image includes: Inputting the cotton field overhead image into a third model to obtain a cotton field overhead image with opened cotton boll positions annotated as output by the third model; wherein the third model is trained using the cotton field overhead image as sample data and the corresponding annotated opened cotton boll positions as label data; The marked position of the cotton bolls in the cotton field overhead image is used as the common position of the cotton bolls in the cotton field overhead image and the cotton field simulation image.
5. The method for remote sensing monitoring of cotton field boll opening rate based on neural network according to claim 2, characterized in that: The steps of generating a cotton field top view by the CottonXL model include: Acquiring growth information of cotton in a cotton field; the growth information includes cotton variety, sowing time, planting density, plant row configuration, topping time, number of cotton fruiting branches, and meteorological data throughout the growth period; Setting parameters of the CottonXL model according to the growth information to obtain a parameter-corrected CottonXL model; and using the parameter-corrected CottonXL model to output a top view of the cotton field according to the growth information. The CottonXL model after parameter correction outputs a cotton field bird's-eye view that meets preset conditions; wherein the preset conditions include the number and type of the cotton field bird's-eye view.
6. The method for remote sensing monitoring of cotton field boll opening rate based on neural network according to claim 1, characterized in that: Determining the target relationship function includes: Output the number of opened cotton bolls in the cotton field top view by using the CottonXL model, and calculate the density of opened cotton bolls in the cotton field top view; The target relationship function is determined based on the corresponding relationship between the cotton boll opening density and the cotton field opening rate in the cotton field top view output by the CottonXL model.
7. A remote sensing monitoring device for cotton field boll opening rate based on neural network, characterized in that: include: The acquisition module is used to obtain the current cotton field image collected by the drone; A monitoring module is configured to input the current cotton field image into a first model to obtain the number of opened cotton bolls in the current cotton field image output by the first model; wherein the first model is trained using a first historical cotton field image as sample data and the corresponding annotated number of opened cotton bolls as label data; A first determining module is configured to determine a density of opened cotton bolls in the current cotton field image based on the number of opened cotton bolls; The second determining module is configured to determine the cotton field boll opening rate based on the boll opening density and a predetermined target relationship function; the target relationship function is a relationship function between the boll opening density and the cotton field boll opening rate.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the neural network-based remote sensing monitoring method for cotton field boll opening rate according to any one of claims 1 to 6 is implemented.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the neural network-based remote sensing monitoring method for cotton field boll opening rate according to any one of claims 1 to 6 is implemented.
10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the neural network-based remote sensing monitoring method for cotton field boll opening rate according to any one of claims 1 to 6 is implemented.