Adaptive Enhancement Network With Border Injection for Animal Parasite Eggs Detection

Abstract

Accurate detection of parasitic eggs can help veterinarians detect issues in the early stages of infection, ensuring timely and effective treatment. However, existing methods are prone to losing the critical location information of the target parasites, resulting in detection errors. In this paper, we propose an adaptive enhancement network with border injection (BIAE-Net) to improve the feature extraction of different types of parasite eggs and enhanced detection efficiency. We introduced a border stream of target eggs to explicitly supervise edge information and significantly enhance the model’s ability to perceive target locations. The Dense Transformer Module was designed to extract detailed global information of the target to minimize the loss of critical features, and the Adaptive Border Attention Module was introduced to capture spatial information. The proposed BIAE-Net was rigorously tested on a sheep dataset consisting of 1344 images, outperforming the leading YOLOv6 model with improvements in Precision, Recall, and mAP by $4.32~\%$ , $4.45~\%$ , and $3.7~\%$ , respectively. Additionally, we demonstrated the efficiency of the model by testing it on a collected dataset of human hookworm eggs using transfer learning. Accurate detection helps to determine the most suitable treatment plan for the effective clearance or control of parasitic infections and improves the treatment success rate.

Keywords

• Target detection
• border injection
• dense transformer
• parasite eggs