An Improved Deep Residual Convolutional Neural Network For Plant Leaf Disease Detection

Abstract

Propose: When they are growing, plants are vulnerable to an assortment of diseases. One of the trickiest issues in agriculture is the early diagnosis of plant diseases. To raise the standard of crop cultivation, it is crucial for one to identify and diagnose crop leaf diseases. The machine-learning identification of crop disease of leaves using machine learning technology offers outstanding accuracy and no subjective judgement mistake as compared to the traditional human detection strategy.

Aims: Compare existing cutting-edge models for the detection of plant leaf disease with the suggested enhanced deep residual CNN.

Design/Methodology: ResNet197 was created using six blocks of levels. Using an integrated disease of plants picture dataset, ResNet197 was programmed and evaluated. The added data of the plant's leaf disease picture recuperation was created using the following techniques: expanding, trimming, flipping, cushioning, rotation, quadratic transformation, the saturation point, and hue conversion. 

Results: The dataset comprised 154,500 photos of both wholesome and ill plant leaves, as well as 103 picture classes of 22 different plants that were simultaneously sick and healthy. The ResNet197 the hyper parameter settings and layers were adjusted using the method of evolutionary search approach. 

Conclusion: On the test dataset, it yielded an average precision for classification of 99.58 %. The experimental outcomes surpassed both current transfer learning methods and Res Net configurations.