AUTHOR=You Jie , Jiang Kan , Lee Joonwhoan 

TITLE=Deep Metric Learning-Based Strawberry Disease Detection With Unknowns

JOURNAL=Frontiers in Plant Science

VOLUME=Volume 13 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2022.891785

DOI=10.3389/fpls.2022.891785

ISSN=1664-462X

ABSTRACT=There has been a lot of research and large progress on plant disease detection based on deep object detection models. However, with unknown diseases, it is hard to find a practical solution for plant disease detection. This paper proposes a simple but effective strawberry disease detection scheme with unknown diseases that can provide applicable performance in real field. In the proposed scheme, the known strawberry diseases are detected with deep metric learning (DML)-based classifiers along with the unknown diseases that have certain symptoms. The pipeline of our proposed scheme consists of two stages; the first is object detection with known disease classes, while the second is a DML-based post-filtering stage. The second stage has two different types of classifiers, one is softmax classifiers for only known diseases, and the other is K-nearest neighbor (K-NN) classifier for both known and unknown diseases. In the training of the first stage and the DML-based softmax classifier, we only use the known samples of strawberry disease. Then, the known (a priori) unknown training samples are included to construct K-NN classifier. The final decisions for known diseases are made from the combined results of the two classifiers, while unknowns are detected from the K-NN classifier. The experimental results show that the DML-based post-filter is effective at improving the performance of known disease detection in terms of mAP. Also, the separate DML-based KNN classifier provides high recall and precision for known, and known unknown diseases as amounts of 96.6 % and 97.7 % in average, respectively, that could be exploited as a Region of Interest (ROI) classifier. For the real field data, the proposed scheme achieves the high mAP of 93.7 % to detect 7 classes of strawberry disease, and reasonable result for unknowns. This implies that the proposed scheme can be applied to find disease-like symptoms due to real known and unknown diseases or disorders for any kind of plant.