Recognition of diseased Pinus trees in UAV images using deep learning and AdaBoost classifier

Abstract

Recognition of diseased Pinus trees in unmanned aerial vehicle (UAV) images is beneficial to the dynamic monitoring and control of Pinus tree diseases in large areas. However, the low resolution and complex backgrounds of UAV images limit the accuracy of traditional machine learning methods in recognising diseased Pinus trees. This study presents a method for recognising diseased Pinus trees that combines deep convolutional neural networks (DCNNs), deep convolutional generative adversarial networks (DCGANs), and an AdaBoost classifier. DCGANs can expand the number of samples of diseased Pinus trees to solve the problem of insufficient training samples. DCNNs are used to remove fields, soils, roads, and rocks in images to reduce the impact of complex backgrounds on target recognition. The AdaBoost classifier distinguishes diseased Pinus trees from healthy Pinus trees and identifies shadows in background removal images. Experimental results show that the proposed method has better recognition performance than K-means clustering, support vector machine, AdaBoost classifier, backpropagation neural networks, Alexnet, VGG, and Inception_v3 networks.