Deep Network for Smart Precision Agriculture through Segmentation and classification of Crops

Abstract

Agriculture meets two of humanity's most fundamental needs, including food and fiber. Advancements in different farming techniques have paced up the agricultural field in the past century. However, the growing population and rising food demand require advancements in technology to fulfill the demand for food and agricultural products, meeting future food demands at a lower cost. Precision agriculture is gaining popularity due to its efficient output through high-tech machinery and robots. In this paper, our primary focus is on Smart Precision Agriculture (SPA) for segmentation and classification of crops and weeds through data augmentation technique such as training of network through image patches. Initially, we divided the images into patches and normalized them. Then, these patches were fed to Convolutional Neural Networks (CNN) to generate the pixel-wise segmentation of crops and weeds. Furthermore, we extracted region of interest (ROI) vegetation blobs by using the segmentation results. Finally, these ROIs are classified in crops and weeds using another network. We used the Carrot/Weed image and Sugar Beets datasets to perform experiments. The proposed system showed superiority in the experimental results.