Image Processing with Convolutional Neural Networks for Classification of Plant Diseases

Convolutional neural networks (CNNs) are one of the most promising techniques from computer vision that can generate substantial gains in the most varied classification problems. In that sense, this paper aims to perform the image processing on XDB plant disease database for enhancing the classification of 20 different diseases from 10 distinct plant species. This processing encompassed two pre-processing activities (image selection and resizing) and the application of two modified VGG architectures, VGG16 and VGG19, along with pre-trained weights from ImageNet database. A comparison study was carried out based on classification metrics, such as accuracy, precision, recall and F1-score. The obtained results demonstrated that for this particular database, a pre-trained CNN with depth equal or smaller than VGG16 can compute disease-sensitive features which aggregate more refinement to recognize plant pathologies, which justified the better performance of VGG16 against VGG19 on training, validation, and test data.