Indian Flora Project: Social Image Data-Based Plant Species Identification and Disease Detection using Deep Learning
Plant species data collection and identification is a crucial step in the conservation of biodiversity and sustainable productivity of agriculture. Speeding up this process can hasten the efforts taken towards the protection of plant species and help in educating the public. One of the significant contributors to an economy, agricultural productivity, is hugely affected by plant leaf diseases. Automatic detection of leaf diseases at an early stage is the need of the hour, in order to eliminate the traditional and highly unprecise naked-eye predictions made by experts. This paper proposes a collaborative workflow for image-based plant identification as a way to engage new contributors and provide botanical data to the public. At the time of writing, an image database of 100 Indian species has been manually collected. This initial database has been synchronized with growing data from a collaborative web-portal and mobile application, where users can add new observations and even query for species identification. We employ a convolutional neural network model (ResNet) to automatically identify 68 of the collected plant species. Experimental observations using the proposed approach shows efficient computation time and high top-5 precision of 99.85% compared to the state-of-the-art approaches that focus on hand-engineered features for detection.
Keywords: Plant Classification, Leaf Recognition, Collaborative Network, Leaf Disease, Crop Diseases, Disease Detection, Deep Learning, Convolutional Neural Network, Fine-Grained Image Classification