CN-121982583-A - Early identification method for sea buckthorn fruit fly pest on basis of multispectral imaging of unmanned aerial vehicle

CN121982583ACN 121982583 ACN121982583 ACN 121982583ACN-121982583-A

Abstract

The invention relates to the technical field of agricultural disaster monitoring, in particular to a sea buckthorn fruit fly pest early identification method based on unmanned aerial vehicle multispectral imaging. The method comprises the steps of collecting flight data in a key weather period of sea-buckthorn through an unmanned aerial vehicle multispectral imaging system, preprocessing and correcting radiation of image data, extracting sea-buckthorn canopy areas and calculating spectral characteristics of the sea-buckthorn canopy areas, constructing and training an improved deep convolution neural network model which integrates a multiscale characteristic fusion structure and a channel attention mechanism, and can deeply analyze spatial modes and internal correlations of the spectral characteristics, and finally utilizing the model to realize early identification and positioning of sea-buckthorn around fruit fly insect pests. The invention can effectively capture early weak spectral response change caused by insect damage, and improves the automation degree, timeliness and accuracy of the identification operation.

Inventors

ZHANG ZE
CHEN XIANGYU
ZHANG QIANG
LI YU
CHEN SHUN
Qin Shizhe

Assignees

石河子大学

Dates

Publication Date: 20260505
Application Date: 20260109

Claims (10)

1. The early identification method for the sea buckthorn fruit fly insect pests based on the multispectral imaging of the unmanned aerial vehicle is characterized by comprising the following steps: s1, performing flying operation above a sea-buckthorn planting area through a multi-spectral imaging system carried by an unmanned aerial vehicle, and collecting multi-spectral image data of sea-buckthorn canopy, wherein the multi-spectral image data comprises at least 5 discrete wave bands in a wavelength range from 400nm to 1000nm; s2, preprocessing the collected multispectral image data, including radiation correction and geometric correction, so as to eliminate the influence of environmental illumination factors and image distortion and obtain standardized earth surface reflectivity data; s3, extracting a sea-buckthorn canopy region from the preprocessed multispectral image data, and calculating spectral characteristic parameters of the region, wherein the spectral characteristic parameters comprise reflectivity ratios of at least 3 different wave bands and 2 vegetation indexes; S4, inputting the extracted spectral characteristic parameters into a pre-trained pest identification model, wherein the pest identification model adopts an improved deep convolutional neural network architecture, and outputting a pest identification result by analyzing a spatial distribution mode of the spectral characteristic parameters; S5, generating an early identification report of the sea buckthorn fruit fly insect pests according to the model output result, wherein the report contains insect pest occurrence position and severity information.
2. The early identification method of the sea buckthorn fruit fly pest on the basis of the multispectral imaging of the unmanned aerial vehicle, which is disclosed in claim 1, is characterized in that the multispectral imaging system in the step S1 comprises at least 5 discrete wave bands, namely a blue wave band of 450nm plus or minus 5nm, a green wave band of 550nm plus or minus 5nm, a red wave band of 650nm plus or minus 5nm, a red wave band of 720nm plus or minus 5nm and a near infrared wave band of 800nm plus or minus 5 nm.
3. The early identification method of the sea buckthorn fruit fly pest on the basis of the multispectral imaging of the unmanned aerial vehicle as claimed in claim 1, wherein the radiation correction in the step S2 comprises the following substeps: s21, converting an original digital value into a surface reflectivity through an atmosphere correction model by utilizing illumination intensity data acquired by the unmanned aerial vehicle at the same time; S22, carrying out reflectivity standardization processing based on standard whiteboard reflectivity data; And S23, eliminating sensor noise influence by using a dark current correction method.
4. The early identification method of the sea buckthorn fruit fly pest on the basis of the multispectral imaging of the unmanned aerial vehicle as set forth in claim 1, wherein the calculation of the vegetation index in the step S3 adopts the following formula: normalized difference vegetation index ndvi= (R800R 650)/(r800+r650), Photochemical reflectance index pri= (R531R 570)/(r531+r570), Where R represents the reflectance value of the corresponding band.
5. The early identification method of the sea buckthorn fruit fly pest on the basis of the multispectral imaging of the unmanned aerial vehicle as claimed in claim 1, wherein the improved deep convolutional neural network architecture in the step S4 comprises the following steps: an input layer for receiving a two-dimensional characteristic diagram composed of multiband spectral characteristic parameters; The feature extraction module comprises 3 continuous convolution layers, and each convolution layer is connected with a batch normalization layer and a ReLU activation function; a spatial attention module for enhancing a characteristic response of the region of interest; and the classification output layer outputs the pest identification result by using a softmax function.
6. The early identification method of the sea buckthorn fruit fly pest on the basis of the multispectral imaging of the unmanned aerial vehicle, which is disclosed in claim 5, is characterized in that the convolution kernel sizes of 3 convolution layers in the feature extraction module are 7×7, 5×5 and 3×3 respectively, the channel numbers are 32, 64 and 128 respectively, and a depth separable convolution structure is adopted to reduce the number of parameters.
7. The method for early identifying sea buckthorn fruit fly pests around a sea buckthorn fruit fly based on multispectral imaging of an unmanned aerial vehicle according to claim 5, wherein the operation of the spatial attention module comprises: carrying out global maximum pooling and global average pooling on the input feature map to obtain two space feature descriptors; after splicing the two feature descriptors, generating a space attention weight graph through a 7×7 convolution layer; And multiplying the space attention weight graph with the original feature graph element by element to obtain a weighted feature graph.
8. The early identification method of the sea buckthorn fruit fly pest on the basis of the multispectral imaging of the unmanned aerial vehicle as set forth in claim 1, wherein the training process of the pest identification model in the step S4 comprises the following steps: s41, collecting a multispectral image data set containing a health sample and insect pest samples with different infection degrees; S42, carrying out data enhancement on the training sample, including random rotation, mirror image overturning and color dithering; s43, solving the problem of unbalanced category in the training data by using a focus loss function; s44, adopting a progressive learning strategy, pre-training on a large-scale plant disease data set, and then fine-tuning on a special data set for the sea buckthorn fruit fly pest.
9. The early identification method of the sea buckthorn fruit fly pest on the basis of the multispectral imaging of the unmanned aerial vehicle according to claim 8, wherein the adjustment factor gamma of the focal point loss function in the step S43 is set to be 2.0, and the weight factor alpha is dynamically adjusted according to the number of various samples.
10. The early identification method of the sea buckthorn fruit fly pest surrounding based on the multispectral imaging of the unmanned aerial vehicle according to claim 1, wherein the generation of the early identification report in the step S5 further comprises the following processing: performing spatial clustering on the recognition result output by the model to eliminate discrete false detection points; mapping the identification result to a sea-buckthorn planting region distribution map by combining geographic information system data; and 3 grades are classified according to the severity of insect pests, and different colors are used for visual display.

Description

Early identification method for sea buckthorn fruit fly pest on basis of multispectral imaging of unmanned aerial vehicle Technical Field The invention relates to the technical field of agricultural disaster monitoring, in particular to a sea buckthorn fruit fly pest surrounding early identification method based on unmanned aerial vehicle multispectral imaging. Background Sea buckthorn is used as an important ecological economic tree species, and fruits of sea buckthorn are often seriously threatened by the insect damage of around fruit flies. The insect pest has the characteristics of strong concealment and difficult visual perception at the initial occurrence stage, but if the insect pest is not recognized and controlled in time, the quality of fruits is reduced, the yield is reduced sharply, and the economic loss is caused. Therefore, the early-stage accurate identification technology capable of realizing large-area and high-efficiency sea buckthorn fruit fly pest around has important practical significance for guaranteeing healthy development of sea buckthorn industry, reducing pesticide abuse and protecting ecological environment. At present, the monitoring of the insect pest mainly depends on manual field investigation, and the method is low in efficiency, limited in coverage, highly dependent on experience of a practitioner and easy to miss in the early stage of insect pest occurrence. With the development of remote sensing technology, part of researches start to try to carry out crop health monitoring by using a multispectral camera carried by an unmanned plane. However, in the prior art, a wide-band vegetation index designed for general vegetation health evaluation is directly adopted, and the indexes have insufficient specificity on early and weak physiological stress response caused by specific insect pests of the sea buckthorn fruit fly, are easy to be interfered by factors such as normal growth change of plants, environmental background fluctuation and the like, so that the identification accuracy and timeliness are not ideal. In addition, most methods rely on single-phase image analysis, lack effective capture of dynamic development processes of insect pest stress, and difficulty in providing reliable early warning information before irreversible damage is caused by insect pests. Therefore, the early identification method for the sea buckthorn fruit fly pest on the basis of the multispectral imaging of the unmanned aerial vehicle is provided for solving the defects in the prior art, and the core to be solved is how to overcome the defects that the traditional method and the existing remote sensing technology are insensitive to weak stress information in the early stage of the sea buckthorn fruit fly pest, have poor specificity and are easy to be interfered. Disclosure of Invention In order to overcome the defects in the prior art, the embodiment of the invention provides an early identification method for sea buckthorn fruit fly pest around based on unmanned aerial vehicle multispectral imaging, so as to solve the problems in the prior art. In order to achieve the purpose, the invention provides the following technical scheme that the early identification method for the sea buckthorn fruit fly pest on the basis of unmanned aerial vehicle multispectral imaging is characterized by comprising the following steps: s1, performing flying operation above a sea-buckthorn planting area through a multi-spectral imaging system carried by an unmanned aerial vehicle, and collecting multi-spectral image data of sea-buckthorn canopy, wherein the multi-spectral image data comprises at least 5 discrete wave bands in a wavelength range from 400nm to 1000nm; s2, preprocessing the collected multispectral image data, including radiation correction and geometric correction, so as to eliminate the influence of environmental illumination factors and image distortion and obtain standardized earth surface reflectivity data; s3, extracting a sea-buckthorn canopy region from the preprocessed multispectral image data, and calculating spectral characteristic parameters of the region, wherein the spectral characteristic parameters comprise reflectivity ratios of at least 3 different wave bands and 2 vegetation indexes; S4, inputting the extracted spectral characteristic parameters into a pre-trained pest identification model, wherein the pest identification model adopts an improved deep convolutional neural network architecture, and outputting a pest identification result by analyzing a spatial distribution mode of the spectral characteristic parameters; S5, generating an early identification report of the sea buckthorn fruit fly insect pests according to the model output result, wherein the report contains insect pest occurrence position and severity information. Preferably, the multispectral imaging system in step S1 comprises at least 5 discrete bands, specifically a blue band of 450nm plus or minus 5nm, a green band of