Vision-Based Plant Disease Detection System Using Transfer and Deep Learning

Abstract.

We have developed a vision-based system to detect symptoms of leaf scorch on leaves of Olea europaea L. infected by Xylella fastidiosa. Previous works predicted disease from leaf images with deep learning but required a vast amount of data. Crowd sourcing generated the required amount of data (PlantVillage project), but this has limited applicability to commercial species that are inaccessible for laypeople to photograph and diseases that require an expert to detect. In this paper, we demonstrate that transfer learning can be leveraged when it is not possible to collect thousands of new leaf images. Transfer learning is the re-application of an already trained deep learner to a new problem. We present a novel algorithm to fuse the data at different levels of abstraction to improve convergence of transfer learning. The algorithm discovers low-level features from raw data to automatically detect veins and colors that lead to symptomatic leaves. Leaf scorch is detected with a true positive rate of 98.60 ± 1.47%. Experiment included 100 control, 99 leaf scorch and 100 abiotic stress images. Results were obtained with a convolutional neural network trained with the stochastic gradient descent method. This work shows potential for early, automatic, quantitative detection of disease in plants with reduced diagnosis time and cost.