Transferable Deep Network Driven Medicinal Plant Disease Detection and Classification System

This paper proposes a first of its kind deep learning driven medicinal plants disease detection and classification system. In sync with structural and textural diversities, at first it performed pre-processing like resizing, contrast equalization. Histogram equalization and Z-score normalization. The previously processed datas were then subjected to feature extraction using a popular transferable deep network called AlexNet. With the aid of the ADAM optimizer, the 4096-dimensional deep features that were collected were processed for learning and classification at the Softmax layer, which distinguished between normal and diseased medicinal plants in each input image. The simulation results confirmed that the proposed AlexNet deep network driven solution yields medicinal plant’s disease detection and classification accuracy of 97.27%, precision 95.01%, recall 94.62% and F-Measure 94.81%.