Detection of Weeds Using Image Processing and Deep Learning Techniques

Abstract

Identifying the weeds in the Vegetable Ranch has been becoming difficult because the plants of the crop are spaced randomly. At this point, by doing research, weeds are identified in the crop through traditional approaches, which mainly focus on the direct differentiation of weeds. Detecting the weeds through the naked eye might be difficult because there are enormous varieties of weeds such as narrow-wide leaf and broadleaf in plantings. The automation used for implementing identification of weeds is deep learning (DL) and image processing (IP). Firstly, the convolutional neural network (CNN) algorithm is used to recognize the weeds by drawing the bounding boxes around the green plants, and the leftover parts are identified as crops. Later on, support vector machine (SVM) is used on same dataset, and confusion matrix and accuracy are generated. Agri_data is the dataset used for training and testing data. By using the algorithms, we can identify whether they are weeds or crops. Accuracy of CNN and SVMs is compared for weed identification and prediction.