Weed density classification in rice crop using computer vision

Abstract Weeds are normally sprayed uniformly across field irrespective of their density in the field. Automatic methods can be developed that help us to locate and spray weed according to its density. Automatic weed detection is necessary for precision spraying that can improve rice yield and reduce production cost. This aids farmers to get an idea of local weed coverage and spray only the weed infested areas of field. This paper proposes two classification techniques to distinguish images based on their weed density. The weed covered by our work is of category grass and is called nutgrass in rice. We aim to classify images into three classes according to grass density. The first technique uses texture features extracted from gray level co-occurrence matrix (GLCM) and produces an accuracy of 73% using Radial Basis Function (RBF) kernel in Support Vector Machine (SVM). Another technique proposed uses features such as moments that are invariant to scale and rotation to classify grass density. The second technique outperforms the first one with an accuracy of 86% using Random Forest classifier. A comparison of these two techniques in terms of execution timing is also conducted. These techniques are implemented in MATLAB and tested on a dataset of hundred images that are collected from an actual rice field.