Real-Time Plant Disease Classification and Targeted Remedial Strategies Using Few-Shot Learning in Mobile Applications

Plant diseases affect agriculture most frequently around the world, causing economic losses that can affect food security, particularly in regions where there are few facilities to provide expert diagnosis. Previous research has studied the application of neural network techniques, like convolutional neural network (CNN), transfer learning etc. in plant disease detection, with good prospects of success. However, such strategies are often reliant on large-scale data collection, copious amounts of training and targeting for limited crop types, a world that never existed in reality because diseases were and will be heterogeneous and data limited. In this paper, a few-shot learning (FSL) approach is proposed for the real-time classification of plant diseases for five different major crops: grapes, corn, strawberries, soybeans and sugarcane. This approach can achieve the best performance in detecting the disease while using the smallest training set. The performance of classifiers is assessed using traditional evaluation metrics such as accuracy, sensitivity, specificity, precision and F1 score with the maximum accuracy of 97.8% in the case of grapes. In addition, the model was deployed in mobile application in which the disease classification can be done from camera and gallery images, and management recommendations for the particular crop can be provided. The work introduced in this paper provides a stepping stone to few-shot learning for plant disease detection and lays a foundation for bridging complex AI techniques to real-world, in-field agriculture applications.