Automated abnormal potato plant detection system using deep learning models and portable video cameras

Abstract

Potatoes are the world’s most important root and tuber crop. A diseased seed potato can produce approximately 10 potato tubers, and the disease can propagate through the seed potato production cycle. To promote stable potato production, quality seed potatoes that are healthy and disease-free should be supplied. The Japanese government established a propagation system for the production and distribution of seed potatoes. Experienced laborers are required in the fields for visual inspection and removal of abnormal plants during seed potato production. To aid visual detection, reduce labor effort, and improve assessment time, we developed an automated abnormal potato plant detection system that utilizes a portable video camera and deep learning models. The proposed system detects abnormal plants or leaves considering the stage of growth. It detects three cases: (i) abnormal potato plants in the early growth stage, (ii) abnormal potato plants in comparison to the surrounding plants, and (iii) abnormal potato leaves. For the abnormal and healthy potato plant classification, the accuracy was ~90%, and the average precision (AP) for detection was 78.2%. Furthermore, the classification accuracy of the abnormal and healthy potato leaf classification was 96.7%, and the AP for detection was 90.5%. Therefore, the proposed system can be used to detect abnormal potato plants.