Regional location of plant diseases helps to develop targeted prevention and control strategies to improve crop production efficiency and reduce agricultural losses. Recently, some important advances have been made in computer vision-based methods. However, most of the current mainstream methods are based on target detection or rotation detection, which always include environmental noise in the results of locating disease areas. In this paper, we revisit the task from the perspective of segmentation tasks and propose a feature reinforcement module. Specifically, we effectively focus on the shortcomings of previous methods where neighbourhood features cannot interact by looking at the shortcomings of the Feature Pyramid Network (FPN) in the feature extraction process through multi-scale features repetitive sampling. In addition, we compared the dataset annotation methods under different annotations such as target detection, rotation detection, and segmentation, and the visualisation well demonstrates the need to use segmentation methods. In the final experimental results, our method is proved to improve mAP by 1.3% and mIOU by 1% over state-of-the-art methods. A large number of methods have demonstrated the superiority and reliability of our method.
Crop pest control is one of the important tasks for crop yield. However, multi-class pests and high similarity in appearance bring challenges to precision recognition of pests. In recent years, deep-learning based algorithms in object detection have achieved an excellent result, such as the YOLO detector, which can balance accuracy and speed. YOLO performs well in detecting normal size objects, but has low precision in detecting small objects. The accuracy decreases notably when dealing with pest dataset, which have large-scale changes and multi-class. To solve the detection problem of multi-scale pest, we propose a detector named YOLO-pest based on YOLOv4 to improve the performance of pest detection. Our approach includes using lite but efficient backbone mobileNetv3 and lite fusion feature pyramid network. The improved detector significantly increased accuracy while remaining fast detection speed. Experiments on the constructed Croppest12 dataset show that our improved algorithm outperforms other compared methods.
Computer vision techniques are an important application for intelligent pest detection. However, it suffers from serious problems and challenges, especially distinguishing the targets of pests with high similarity, small size, and sample unbalance. In this paper, a domain-adaptive-calibrated-free-anchor detection network (DACFA-Det) is proposed, in which a balanced learning mechanism is added to the detection network, dealing with sample imbalance, feature similarity confusion, and center point inaccurate. Our method is evaluated on the re-established similar pest dataset (SPD). The final experimental results show that our method can obtain 44.0% mAP (Average Precision) on the SPD. The testing speed achieve 0.045s per image, meeting the real-time requirement, which proved the effectiveness and efficiency of our method for agricultural similar pest detection.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.