Research Area:  Machine Learning
Food security, which has currently attracted much attention, requires minimizing crop damage by timely detection of plant diseases. Therefore, the automatic identification and diagnosis of plant diseases are highly desired in agricultural information. In this paper, we propose a novel approach to identify plant diseases. The method is divided into two parts: starting with the enhancement of the artificial neural network, the extracted pixel values and feature values are input to the enhanced artificial neural network for the image segmentation; then, following the establishment of a CNN based model, the segmented images are input to the proposed CNN model for the image classification. The proposed approach shows an impressive performance in the experimental analyses. It achieved an average accuracy of 93.75% to identify the crop diseases under the complex background conditions, and the validation accuracy was, on average, 10% higher than that of the conventional method. Additionally, almost all the plant disease samples were correctly detected by the proposed approach, and thus the recall rate achieved 100%. The experimental finding presents a substantial performance relative to other state-of-the-art methods and demonstrates the efficiency and extensibility of the proposed approach.
Author(s) Name:  Junde Chen, Jinxiu Chen, Defu Zhang, Y. A. Nanehkaran & Yuandong Sun
Journal name:  Machine Vision and Applications
Publisher name:  Springer
Volume Information:  volume 32, Article number: 31 (2021)
Paper Link:   https://link.springer.com/article/10.1007/s00138-020-01150-w