Transfer Learning based Tomato Leaf Disease Detection for mobile applications

Abstract

An estimated 15-25% of potential crop production in India is lost to pests, diseases and weeds. Advanced technologies for the early detection of crop diseases are required to achieve food security. Convolutional Neural Networks have found large success in vision problems such as classification and object detection. They have been used extensively in a plethora of fields in the recent years including robotics, healthcare and agriculture. However such deep learning approaches are computationally expensive and have large memory and power requirements. The paper aims to use transfer learning to obtain effective results for use on mobile devices at reduced costs using pre-trained EfficientNetB0, MobileNetV2 and VGG 19 models as feature extractors. 15,000 images from 9 types of diseases and one healthy class from the tomato plant are used to show the potential use of such approaches in agricultural applications.