In order to effectively prevent the global spread of malaria, classical deep learning models have been applied to malaria detection. However, these models generally suffer from low accuracy. In order to address the identified limitations, an Efficient Target-Oriented YOLO model (ET-YOLO) is proposed in this thesis. To address the limited discriminability of C3k2 in malaria microscopy images, we redesigned it into C3k2fECA, which integrates efficient channel attention and a refined fusion pathway to emphasize parasite-related regions. We further developed C3k2fTR, leveraging Transformer-based global context modeling to remedy the loss of contextual cues and improve robustness under complex backgrounds. In addition, a lightweight ConvNeXt variant, CNeB (ConvNeXt Block), was incorporated to effectively reduce model parameters while maintaining strong representational capacity. The experimental results of the improved model on two different datasets demonstrate the effectiveness of the improved model, specifically achieving [email protected] of 86.2% and 77.9% on two different datasets, both of which outperform other traditional YOLO models, while the number of parameters is reduced by about 7.2% compared to the reference model. A balance has been achieved between detection accuracy and computational resource utilization, providing a practical technical solution for malaria control in resource-constrained regions.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号