融合CNN多卷积特征与HOG的番茄叶部病害检测算法

A Tomato Leaf Disease Detection Algorithm Based on CNN Multi-convolution Feature and HOG

传统的番茄叶部病害检测依赖于耗时费力的人工特征设计,必须针对病害的不同分类精心设计相关特征。番茄叶部病害症状复杂,人工设计特征较难。卷积神经网络(convolutional neural network,CNN)可以自动挖掘出隐藏在病害图像内部的抽象特征,在图像识别领域性能优越。该研究提出采用CNN与传统的HOG+SVM算法相结合的方法,抽取番茄叶部病害的浅层特征,将其输入到HOG生成HOG特征并合并,最后输入SVM分类器得到病害检测结果。该研究方法能够改进番茄叶部病害的检测精度。

Traditional tomato leaf disease detection relies on time-consuming and laborious artificial feature design and must be carefully designed for different types of tomato diseases.Symptoms of tomato leaf diseases are complex and adopting the methods of artificial design features is difficult.Convolutional Neural Network(CNN)can automatically discover the abstract features hidden in the diseased images,and its performance is superior in the field of image recognition.In this paper,the method combining CNN with traditional HOG+SVM algorithm was proposed to extract the shallow features of tomato leaf diseases,input them into HOG to generate HOG feature and merge them,and finally input them into SVM classifier to obtain disease detection results.Experiments showed that this method could improve the precision of tomato leaf disease detection.