Weight Pruning in Convolutional Neural Network for Efficiency in Plant Disease Detection

Abstract

Plant infection is an industrious issue for smallholder ranchers, representing a danger to their livelihoods and meal preservation. Image classification in agriculture has become possible thanks to the latest spike in mobile use and computer vision models. Convolutional neural networks are the cutting edge in picture acknowledgment, and they can give a quick and precise finding. In spite of the way that these CNN models are profoundly helpful in an assortment of PC vision exercises, the high number of boundaries makes them computational and memory escalated. Pruning is a significant procedure for diminishing the quantities of boundaries in a CNN model by wiping out superfluous loads or channels without bargaining generally speaking accuracy. The effectiveness of a weight pruning in CNN model in detecting crop disease is studied in this paper. The created models, which are available as a Web API, can detect different plant diseases according to selected plant. For training and validating the model, the dataset used is new plant diseases dataset available on kaggle. The suggested approach will obtain far better accuracy than the normal approach, according to validation results. This illustrates CNNs’ technological flexibility in classifying plant diseases and opens the way for AI solutions for smallholders.