【摘要】 针对无人机平台获取的高分辨率可见光松树图像,提出一种结合深度卷积神经网络和Adaboost算法的病害松树识别方法,解决传统机器学习方法识别病害松树精确度不高问题.首先利用卷积神经网络训练病害松树模型再利用训练模型将地物中的田地、裸土及黑影等复杂信息剔除掉,提取病害松树、健康松树及黑影区域的颜色和纹理特征,依据提取的特征在剔除地物干扰项后的决策层使用Adaboost分类器进行目标识别.实验结果表明,该方法相较传统的K-means聚类、支持向量机、Adaboost算法、BP神经网络、VGG(visual geometry group)算法等在识别精确度方面有显著提高.更多还原

【Abstract】 Aiming at the high-resolution visible pine tree image acquired by the UAV platform, a method of disease pine identification combined with deep convolutional neural network and Adaboost algorithm was proposed to solve the problem of low accuracy of traditional machine learning method for identifying diseased pine. Firstly, the convolutional neural network was used to train the diseased pine model and then used the pre-training model to remove the complex information such as fields, bare soil and shadows, and extracted the color and texture features of the diseased pine, healthy pine and shadow areas. The Adaboost classifier was used to identify the disease target according to the extracted features in the decision-making layer after the object interference item was removed. The experimental results showed that the proposed method had significantly higher recognition accuracy than traditional K-means clustering, support vector machine, Adaboost algorithm, BP neural network and VGG(visual geometry group) algorithm.更多还原