SDD: Strawberry Disease Detection Framework Based on Computer Vision Neural Networks

Authors:

Abstract
Identifying and treating strawberry diseases are crucial in farmers’ cultivation practices. However, accurate disease identification remains challenging, including limited available datasets, diversity in disease appearance, and low accuracy in disease detection. To address these challenges, we propose a Strawberry Disease Detection (SDD) framework based on multiple computer vision neural networks. We first build a well-annotated dataset of 1,943 images with 7 categories of strawberry diseases. Four popular deep visual detection models, including YOLOv3-tiny, Faster R-CNN, YOLOv6-m, and YOLOv7-m, are incorporated into the SDD framework. We further enhance the YOLOv7 algorithm by integrating data augmentation techniques to diversify the training data. Our algorithm utilizes the SIoU loss function to locate diseases precisely. Extensive experiments are conducted on real-world datasets to evaluate the performance of all networks within the SDD framework. The experimental results indicate that YOLOv3-tiny exhibits the lowest performance in terms of F1score, Precision, and Recall. Conversely, Faster R-CNN, YOLOv6m, and YOLOv7-m demonstrate superior recognition accuracy, despite their relatively higher Giga Floating Point Operations Per Second (GFLOPs). The proposed Enhanced YOLOv7 achieves the best performance, with an mAP of 90.87% and an F1-score of 0.862, respectively.

This publication has 13 references indexed in Scilit: