ResNet-101 Pretrained Model Based Plant Disease Deep Recognition Using Visual Region Approach
Keywords:
CNN, Deep neural network, Gradient-weighted class activation mapping, Hard triplet generation, Residual networkAbstract
Image processing and deep learning approaches habituated ascertaining stalk disorder identification. The majority of research concentrated on identifying illness using photos of complete leaves. Ahead of anatomize notch of stalk disorder proficiency restricted adopting imitative replica. Sickness perception fidelity ameliorated contemporary analysis. In the existing system, stalk disorder scrutiny explores consistently. ResNet-18 is based on pretrained model features, however higher precision was required. In this project we construct stalk disorder with huge dataset. Based on this dataset, we propose a Modified CNN architecture stalk disorder perception. The weights of all split patches are computed to determine the discriminative level of each patch. Finally, modified Convolutional Neural Network accustomed afore extricate facet. Our research helps stalk disorder identification with image processing.
