Fruit Plants Disease Severity Detection and Pesticide Recommendation Using Deep Learning

Abstract

Agriculture is considered to be essential for keeping economic stability in the world. The condition of the fruit plants affects the phenomenon of cultivation directly,  regard for the quantity and quality of fruits. The infringement of plants should be observed early and correctly judged to prevent the fruit farmers from economical damage,  and to possess a sustainable cultivation. In this work, a deep learning based model for disease detection of fruit plants and the evaluation,  estimation of severity level with the proposing of pesticide solution is developed. CNN categorized a classifier by MobileNetV2 base architecture to discriminate over 23 classes, including healthy and unhealthy fruit leaves like apple, banana, grape, mango, orange and strawberry,  and we find a multi class type of the classification based on the two search modes. The severity levels of the infections are calculated according to the Confidence scores, based on the scores suitable pesticide is recommended. The top visual CNN feature map, Grad-CAM inspired,  is employed on CNN model explained what location command the prediction of CNN.  Based on severity and scourge class, the possible medicine use is also advised to aid farmers.  The MobileNetV2 choice is as a lightweight model, merely takes a few and consumes low time.  Our system achieved a state-of-art accuracy of 98.29% with a precision, recall and F1-score of 0.98.