Plant Leaf Diseases Severity Estimation using Fine-Tuned CNN Models

The quantity and quality of agricultural harvests are both severely impacted by crop diseases. Accurately measuring the severity of a disease is vital because it allows farmers to use the appropriate amount and kind of pesticides on the crops that are threatened. Incorrect estimates of disease severity in plants can lead to wasteful or ineffective use of pesticides, making it a difficult assignment even for researchers and plant pathologists. Recently, there has been a rapid rise in the application of machine vision and deep learning methods to smart farming. To guarantee food security, we need to raise output and improve crop quality, but doing so requires a more precise and innovative method of assessing the severity of the crop disease. This research proposes a transfer-learning based strategy for estimating the severity of diseases on tomato leaves with the use of learned VGG-16 / VGG-19 CNN networks and tested on a hybrid dataset consisting of both images captured in the field with a Canon EOS 1500D camera on a white background and images captured under controlled laboratory conditions from the plant village dataset. In addition, the authors made hyper- adjustments to the hyperparameters of pre-trained CNN models to boost their efficacy. To evaluate the efficacy of finely tuned CNN models, the study uses accuracy and loss measurements over multiple iterations on training and validation datasets. When compared to another CNN model evaluated on the same dataset, VGG-16 was shown to obtain superior classification accuracy (92.46%).