| Title |
Research on a New Deep Learning based Pipeline to Minimize Misrecognition of Apple Diseases and Pests Images |
| Authors |
박수빈(Subin PARK) ; (YIN HELIN) ; (DONG JIN) ; (ZHENG RI) ; 구영현(Yeong Hyeon GU) |
| DOI |
https://doi.org/10.5573/ieie.2024.61.6.63 |
| Keywords |
Deep learning; Object detection; Apple diseases and pests; Misrecognition |
| Abstract |
Recently, damage from pests and diseases to crops is increasing worldwide due to climate change and abnormal weather conditions, and therefore accurate pest diagnosis is essential. Currently, deep learning-based image recognition technology is being applied to diagnostic research using various methods such as image classification, image object detection, and image segmentation. However, there are still cases where diseases are misrecognized as pests and pests are misrecognition as disease. Accordingly, this study proposes a new deep learning-based pipeline that can minimize the misrecognition of pests and diseases images in apple crops. This pipeline comprises three main stages, where the first stage automatically detects Regions of Interest (RoI) indicating pest and disease damage from the original image. In the second step, diseases and pests are identified based on the detected images. In the third step, the identified images are input into a diagnostic model to classify the types of diseases and pests precisely. In this study, a total of 17,067 images were used as the data set, consisting of 6 types of apple crop diseases (8,953 images) and 7 types of pests (8,114 images) taken at actual farms. The experimental results showed that while the existing benchmark pipeline experienced 465 misrecognitions, the proposed pipeline reduced misrecognitions to 207 cases, thus decreasing misrecognitions by 258 cases and demonstrating its significant effect in reducing misrecognitions in pest and disease diagnosis. |