Recognition of Apple Leaf Diseases using Deep Learning and Variances-Controlled Features Reduction

Abstract Agricultural industry plays a significant role in the economy of developing countries by offering several benefits including food, employment opportunities, and income. The prolific plants and agricultural crops are endangered by a variety of pests and plant diseases. About 42% of the world's total agricultural crops are shattered yearly by pests and diseases. Around the world, several computer vision based solutions are adopted which are quite effective in controlling the plant diseases. Despite of large number of available solutions, crops and fruits’ security is not guaranteed, and the applied techniques are unable to provide full prevention from the diseases- leaving the gap for the researchers. In this work, a new deep model is proposed for apple diseases classification. The most common apple diseases are apple scab, brown spot and apple cedar. The infected and healthy images are collected from the plant village dataset to re-train the Inception V3 deep architecture via transfer learning. The extracted features after the transfer learning are later down-sampled using novel variance-controlled approach. Finally, the most discriminant features are classified using the state-of-the-art classifiers to the best accuracy of 97%. Based on the simulation results, a suitable choice of a classifier is also mentioned for this application. A fair comparison with the existing techniques confirms the effectiveness of the proposed technique.