CN-121459225-B - Flue-cured tobacco yield prediction method integrating airborne laser radar point cloud and orthophotos

CN121459225BCN 121459225 BCN121459225 BCN 121459225BCN-121459225-B

Abstract

The application relates to the technical field of remote sensing image crop estimation and provides a flue-cured tobacco yield prediction method integrating airborne laser radar point cloud and an orthographic image, which comprises the steps of acquiring the laser radar LiDAR point cloud and the orthographic image of a measurement area; the method comprises the steps of carrying out tobacco field identification, tobacco Tian Erwei polygon extraction and tobacco field area calculation on an orthographic image by utilizing a pre-trained U-Net semantic segmentation network model, carrying out tobacco field leaf area index extraction based on LiDAR point cloud to further calculate tobacco field average leaf area index, obtaining tobacco leaf surface density value of the tobacco field, and predicting flue-cured tobacco production based on the tobacco field area, the tobacco field average leaf area index and the tobacco leaf surface density value of the tobacco field. The method provides a new method for flue-cured tobacco yield prediction, has the obvious advantages of saving the number of field operators, workload and labor intensity, and can be free from contact, damage, simple and easy to implement, high in precision, high in timeliness and economical and feasible.

Inventors

ZHANG ZHIYONG
TAN XIAOLEI
GAO QIANG
WANG BAOJIAN
DENG ZHICHAO
LIN XIANGGUO
LI JUN
MENG FANCHAO
TIAN LEI
ZONG HAO
WANG LILI
GAO PENGCHENG
YUAN XIN
WU JIAN

Assignees

山东临沂烟草有限公司

Dates

Publication Date: 20260512
Application Date: 20251125

Claims (6)

1. The flue-cured tobacco yield prediction method integrating the airborne laser radar point cloud and the orthographic image is characterized by comprising the following steps of: S1, acquiring a laser radar LiDAR point cloud and an orthophoto of a measurement area; S2, performing tobacco field identification, tobacco Tian Erwei polygon extraction and tobacco field area calculation on the orthographic image by utilizing a pre-trained U-Net semantic segmentation network model; s3, extracting tobacco field leaf area indexes based on LiDAR point clouds, and further calculating tobacco field average leaf area indexes; S4, obtaining a leaf surface density value of tobacco leaves in a tobacco field; S5, predicting the flue-cured tobacco yield based on the tobacco field area, the tobacco field average leaf area index and the tobacco field leaf surface density value; In step S2, the specific method for carrying out tobacco field recognition, tobacco Tian Erwei polygon extraction and tobacco field area calculation on the orthographic image by utilizing the U-Net semantic segmentation network model is as follows: S201, preprocessing an orthographic image, namely cutting an image large image into sub-images with fixed sizes, and normalizing the sub-images; S202, extracting deep semantic features from subgraphs by a U-Net semantic segmentation network model, performing semantic segmentation, outputting a tobacco field probability map of each subgraph, performing binarization processing on the tobacco field probability map, and splicing binarization results of the subgraphs into original orthophoto dimensions; S203, converting a tobacco field connected domain in the spliced binary image into a two-dimensional polygon under geographic coordinates to generate a tobacco Tian Erwei polygon; s204, calculating the area of the two-dimensional polygon; the construction method of the U-Net semantic segmentation network model comprises the following steps: introducing a U-Net model, and then selecting a proper loss function and an optimizer; Improving the encoder structure, and utilizing a coordinate attention and channel space attention module at the network layer jump connection; edge detection scores are introduced.
2. The flue-cured tobacco yield prediction method based on fusion of airborne laser radar point clouds and orthographic images according to claim 1, wherein the method comprises the following steps of: in the step S1, the LiDAR point cloud and the orthographic image need to cover the same area and have the same projection coordinate system.
3. The flue-cured tobacco yield prediction method fusing airborne laser radar point clouds and orthographic images according to claim 2, wherein the method comprises the following steps of: and synchronously acquiring a high-density LiDAR point cloud and true color orthographic images of the measurement area by using the Dajiang Buddhist L2.
4. The flue-cured tobacco yield prediction method based on fusion of airborne laser radar point clouds and orthographic images according to claim 1, wherein the method comprises the following steps of: In the step S3, extracting the tobacco field leaf area index based on the LiDAR point cloud, and further calculating the tobacco field average leaf area index specifically comprises the following steps: S301, cutting out LiDAR point clouds corresponding to each tobacco field based on two-dimensional polygons corresponding to the tobacco field; S302, classifying a tobacco field LiDAR point cloud, firstly classifying an original point cloud into two types of ground points and non-ground points, and then screening tobacco leaf points from the non-ground points; S303, partitioning the classified point cloud of the tobacco field according to a regular grid of 2 x 2m, wherein the number of effective grid blocks is recorded as n, and the block numbers i=0, 1,2, & gt, n-1; S304, calculating the leaf area index of each grid of the tobacco field, wherein the formula is as follows: In the formula, For the leaf area index of the smoke Tian Di i grids, The average scanning angle of all laser radar points of the grid is given by radian value, Is the total number of tobacco leaf points of the grid, The number of all laser radar points in the grid; S305, calculating the average leaf area index of the tobacco field, wherein the formula is as follows: traversing each tobacco field polygon, and calculating the average leaf area index corresponding to the tobacco field according to the methods from step S301 to step S305 。
5. The flue-cured tobacco yield prediction method based on fusion of airborne laser radar point clouds and orthographic images according to claim 1, wherein the method comprises the following steps of: In the step S5, the specific method for predicting the flue-cured tobacco yield based on the tobacco field area, the tobacco field average leaf area index and the tobacco field leaf surface density value is as follows: wherein A is the area of the tobacco field to be measured, the unit is square meter, The average leaf area index of the tobacco field to be measured, the average leaf surface density of the tobacco field to be measured, is expressed as g/square meter, The proportion of the tobacco leaves which are already picked in the tobacco field is distributed between 0.0 and 1.0, The field loss rate is distributed between 0.0 and 1.0, The ratio of the baking loss is distributed between 0.0 and 1.0, 0.001 is a mass unit conversion coefficient, and the unit of P is converted from g to kg.
6. The method for predicting the flue-cured tobacco yield by fusing airborne laser radar point clouds and orthographic images according to claim 5, wherein the method comprises the following steps of: Loss rate in field The value is 0.05, and the ratio of the baking loss is The value is 0.10.

Description

Flue-cured tobacco yield prediction method integrating airborne laser radar point cloud and orthophotos Technical Field The application belongs to the technical field of crop estimation of remote sensing images, and particularly relates to a flue-cured tobacco yield prediction method integrating airborne laser radar point clouds and orthographic images. Background In the tobacco planting process, tobacco seedlings are cultivated according to planting planning, tobacco planting farmers are selected to sign contracts with tobacco farmers in advance, tobacco planting shares are reasonably distributed, land areas are manually measured and positions of the land areas are recorded, and in the tobacco planting stage, the tobacco farmers develop agro-work activities under the guidance of full-time technicians, but face the following problems: (1) The area measurement workload is large, the measurement accuracy is difficult to promote and monitor, the tobacco field area measurement is an important means for monitoring the quantity and the area of tobacco actually transplanted by smokers, the quantity and the area of tobacco actually transplanted by the smokers are consistent with the quantity and the area of tobacco planted by contracts, the area measurement work requires numbering land parcels, the geographic coordinates of the land parcels are measured, the plan view of tobacco planted land parcels is manufactured, the work of area measurement is particularly difficult due to the large tobacco planting range and the limited staff, the accuracy of the area measurement is not high, and in addition, the tobacco field area monitoring is difficult for the part of excessive planting outside the contracts, the monitoring workload is large and the efficiency is low mainly by inspection. (2) The influence factors of the yield prediction are more, in the tobacco growth process, tobacco is not produced or produced is reduced due to possible climate, plant diseases and insect pests and human factors, the tobacco yield is directly influenced, the report of the artificial disaster is inevitably delayed by information, and the disaster receiving degree and the area of the disaster area are difficult to accurately measure manually. (3) The yield prediction is difficult, and in order to reasonably arrange the subsequent processing and manufacturing of tobacco leaves, tobacco yield estimation needs to be carried out, so that the problem of how to quickly, economically and accurately measure and calculate the large-area yield is a great problem. With the development of remote sensing technology, the method for predicting the yield of remote sensing crops gradually takes an important role, and research on crop estimation by students at home and abroad by using the remote sensing technology is mainly focused on crops such as rice, corn, wheat and the like, and has less research on tobacco. Disclosure of Invention The application provides a flue-cured tobacco yield prediction method integrating airborne laser radar point cloud and orthophotos, which aims to solve or partially solve the problems in the background technology. The application provides a flue-cured tobacco yield prediction method fusing airborne laser radar point cloud and orthophotos, which comprises the following steps: S1, acquiring a laser radar LiDAR point cloud and an orthophoto of a measurement area; S2, performing tobacco field identification, tobacco Tian Erwei polygon extraction and tobacco field area calculation on the orthographic image by utilizing a pre-trained U-Net semantic segmentation network model; s3, extracting tobacco field leaf area indexes based on LiDAR point clouds, and further calculating tobacco field average leaf area indexes; S4, obtaining a leaf surface density value of tobacco leaves in a tobacco field; S5, predicting the flue-cured tobacco yield based on the tobacco field area, the tobacco field average leaf area index and the tobacco field leaf surface density value. Preferably, in the step S1, the LiDAR point cloud and the orthographic image need to cover the same area and have the same projection coordinate system. Preferably, the high-density LiDAR point cloud and true color orthographic images of the measurement area are synchronously acquired by using the Xinddhist L2 in the Xinjiang. Preferably, in step S2, the specific method for performing tobacco field recognition, tobacco Tian Erwei polygon extraction and tobacco field area calculation on the orthographic image by using the U-Net semantic segmentation network model is as follows: S201, preprocessing an orthographic image, namely cutting an image large image into sub-images with fixed sizes, and normalizing the sub-images; S202, extracting deep semantic features from subgraphs by a U-Net semantic segmentation network model, performing semantic segmentation, outputting a tobacco field probability map of each subgraph, performing binarization processing on the tobacco field probabil