Autoencoders for semantic segmentation of rice fungal diseases
¹Plarium Inc., 75/1 Uralskaya Str., RU350001 Krasnodar, Russia
²Kuban State Agrarian University named after I.T. Trubilin, 13 Kalinina Str., RU350044 Krasnodar, Russia
³Kuban State University, 149 Stavropolskaya Str., RU350040 Krasnodar, Russia
⁴All-Russian Research Institute of Biological Plant Protection, 1 VNIIBZR Str., RU350039, Krasnodar, Russia
In the article, the authors examine the possibility of automatic localization of rice fungal infections using modern methods of computer vision. The authors consider a new approach based on the use of autoencoders – special neural network architectures. This approach makes it possible to detect areas on rice leaves affected by a particular disease. The authors demonstrate that the autoencoder can be trained to remove affected areas from the image. In some cases, this allows one to clearly highlight the affected area by comparing the resulting image with the original one. Therefore, modern architectures of convolutional autoencoders provide quite acceptable visual quality of detection.