基于深度特征的稻田苗期杂草识别方法研究

Weeds Recognition at Seedling Stage in Paddy Fields Based on Deep Feature

稻田杂草主要在水稻秧苗封行前与其竞争水肥光等资源,也为病虫害提供滋生条件。由于目前主要防控方式为除草剂无选择性地喷施,造成大量的农药浪费和环境污染。由于除草剂针对杂草的靶向喷施可大量减少农药使用量,笔者提出了一种基于深度卷积特征的稻田苗期杂草识别方法,采用支持向量机(Support Vector Machine,SVM)和k最近邻算法(K-Nearest Neighbor Classifier,KNN)两种算法针对6种杂草的深度卷积特征进行识别。实验结果表明:基于SVM深度特征分类准确度高于KNN算法,两种算法针对苗期杂草深度特征识别精度都高于94%,可满足田间杂草除草剂靶向喷施的应用需求。

Weeds at seedling stage in paddy fields is the leading factor to the poor yield and decreased quality, which competing for moisture, nutrients, and light in the paddy field. The strategy of herbicide chemicals is to prevent and control weeds often results in the excessive application of pesticide. Targeted spraying can greatly reduce bag use herbicides without impairing the prevention and control of the weed. This paper presented a new method for weeds recognition at seedling stage in paddy fields based on deep convolution feature. The two algorithms are implemented using deep convolution feature with six weedy plants based on support vector machine(SVM) and K-Nearest Neighbor Classifier(KNN). The results indicated that accuracy of classification with SVM were higher than those of KNN, and identify precision of both algorithms were higher than 94%. The experimental results demonstrate that the proposed method can meet the needs of targeted spraying with weeds at seedling stage in paddy fields.