Plant Phenomics with Deep Learning
Plant disease is a major limiting factor in food production. To prevent production loss, current plant breeding strategies rely on the disease severity rating of a variety of plants. However, traditional manual rating of plants is a low throughput process which necessitates the development of an automatic framework for disease severity rating based on plant images. Since these images are impaired by complex backgrounds, uneven lighting, and densely overlapping leaves, the state-of-the-art frameworks formulate the processing pipeline as a dichotomy problem (i.e. presence/absence of disease), thereby missing crucial information such as accurate disease localization and quantification.
Our researcher, Swati, and her team has been working in this area extensively. To overcome the limitation of traditional methods, she has developed a novel deep learning-based framework. This framework permits simultaneous segmentation of individual leaf instances and corresponding diseased regions using a unified feature map with a multi-task loss function for end-to-end training. We test the framework on a field maize dataset with Northern Leaf Blight (NLB) disease and the experimental results show a disease severity correlation of 73% with the manual ground truth data and run-time efficiency of 5fps.
The following video presentation discusses the details of the problem and solution.
Further details of her work are available in the paper Automatic Quantification of Plant Disease from Field Image Data Using Deep Learning published in WACV 2021.
