Improving Crop Management with Convolutional Neural Networks for Binary and Multiclass Weed Recognition

Weeding is a pre-harvesting process that removes unwanted plants from a land to achieve high agricultural production. Weed management will have an economic impact on both farmers and end users. The primary step of weed control is accurate classification of crops and weeds in the agricultural field. The sooner a new breed of weeds is recognized; the more successful control measures can be taken. Various Machine Learning (ML) and Deep Learning (DL)-based techniques are developed by researchers, for classification of different weeds and crops. Although the pre-trained CNN algorithms work effectively, the wide range of associated parameters makes them computationally intensive. A simplified CNN model with eight hidden layers is proposed in this research work. The proposed light-weight algorithm outperforms the pre-trained models (VGG16, Exception, ResN et152V2, Inception V3)on the publicly accessible corn and radish dataset, with 97.66% and 98.75% accuracy.