<p>In the algae image analysis task, the original underwater red tide microscopic images generally have problems of blurred details and background interference. In order to improve image clarity and achieve complete segmentation of algae, we propose a multi-module collaborative enhancement and segmentation method. In the enhancement stage, we use a cascade of white balance, improved gamma correction, contrast enhancement, and homomorphic filtering, combine with the adaptive weighted particle swarm optimization (AWPSO) algorithm and wavelet fusion technology to effectively correct the color of the algae, enhance the detailed features, and improve the accuracy of edge detection and key point matching. In the segmentation stage, we propose a multi-scale attention mechanism algae segmentation network (AlgaeNet) based on UNet. In the shallow semantic branch, we build the Swin-MLP attention module based on Swin Transformer, and effectively capture the global context information through the sliding window mechanism. In the deep semantic branch, we inherit the UNet architecture and introduce the multi-scale attention module (MSAM) based on the VGG16 encoder. This module integrates the adaptive residual attention layer (ARAL) and realizes the dynamic enhancement of local features through the cascade of channel and position attention. Experiments show that the proposed method significantly improves the segmentation accuracy on the enhanced algae dataset.</p>