LBEF RESEARCH JOURNAL OF SCIENCE, TECHNOLOGY AND MANAGEMENT

Author(s):Sanskar Singh, Dr. Swati Sah

Abstract:Nepal is an agricultural country, where two-thirds of the population depend upon for livelihood. Pesticides and fertilizers play a key role in increasing the productivity of the crops. However, human inability to distinguish the type of plant disease on their own is leading to improper pesticide management. Previous research suggests, about 15-25% of the food products are lost annually whether in pre- or post-harvest activities reason being plant diseases due to improper pesticide management. This project aims to combine modern technologies and easy internet access to develop a tool for improving pesticide management by identifying and classifying plant diseases at farmers level. This paper presents a comprehensive study and development done on recognition and classification of plant leaf using Convolutional Neural Network. For understanding the requirements, study on several traditional approaches along with modern solutions were done during which gaps and possible improvements were found out. Field study was done to identify most common crops at the local level. Dataset on most used crops such as maize, potato, tomato and pepper were used. Each of around 1000 images on diseased and non-diseased leaves from Kaggle were used. Altogether around 10,000 images have been used for the training and validation. After using various CNN architectures, an architecture to best fit for our dataset was created. The custom architecture was able to generate an accuracy of 96%.

Keywords:Plant disease, Pesticide Management, Artificial Intelligence, Rural, Convolutional Neural Network, AI, Agriculture