Identification of leek diseases based on deep learning algorithms

Vegetable diseases and pests continue to be a major threat to food security. Manual identification is time-consuming and labor-intensive due to the wide variety of diseases and pests, and it is also prone to errors. In this paper, a novel and integrated deep learning framework based on convolutional neural networks was proposed for leek disease identification using an improved Cycle-GAN network, gated recurrent units, and recurrent convolutional auto-encoders. We collected healthy and diseased leek leaves under realistic conditions for the experimental implementation. We concentrated on the identification of five leek diseases: virus, rust, blight, sclerotium rolfsii, and botrytis cinerea. It is discovered that the proposed method’s evaluation metrics produce an accuracy of 99.58%, a specificity of 99.62%, a sensitivity of 97.96%, a precision of 99.77%, and an f1-score of 98.19%, which is significantly higher than other cutting-edge deep learning models such as ShuffleNet-V3, Inception-V4, and Efficient-B7. Many aspects of the proposed model implementation provide competitive advantages, such as high recognition precision, strong adaptive ability, and good generalization. In conclusion, the proposed system can accurately classify leek diseases, which is very beneficial in increasing the efficiency of plant pathology image analysis.