Encoder–Decoder Architecture for Crop-Weed Classification Using Pixel-Wise Labelling

Abstract

Agricultural growth is an important pathway in development of any country. Its productivity contributes in full filling the basic need of the human society. The productivity is therefore must be smoothen to provide quality and quantity. Reduction in usage of chemicals like pesticides and herbicides to provide quality and quantity. The major factor that affect the quantity is the presences of weed in the crop field. The nutrient present in the soil is therefore observed by both the weed and the crop. Manual removal of weed from crop is tedious, time consuming and costly. Spraying of herbicides over the field affects the quality of the crop. Emergence of technology in the agriculture field paves the path for selective spraying and robot removal of weed, which requires high accuracy classification of crop from weed. Therefore an Encoder-Decoder architecture based on VGG16 architecture is used for the pixel-wise segmentation. The architecture consists of convolutional layers with ReLU, Normalization layer and max-pooling layer.