A portable intelligent analysis platform for automated microalgae detection and recognition

Abstract

To address the limitations of conventional microalgae identification and detection methods, including difficulties in on-site analysis, reliance on manual microscopic examination, and limited performance of existing intelligent detection algorithms, this study proposes a portable intelligent prototype system for microalgae cell recognition and detection, enabling low-cost and high-precision automated analysis of microalgae cells. First, to overcome the limitations of laboratory-based microscopy, a portable and cost-effective microscopic imaging prototype was developed by integrating optical microscopy principles with CMOS sensing technology, enabling high-resolution imaging of micrometer-scale cells. Subsequently, to achieve accurate and intelligent microalgae cell detection, a novel detection framework, termed LMCF-AlgaeDet, was proposed based on YOLOv8n. To address the challenges associated with the small size, complex morphology, and significant scale variations of microalgae cells, a Global-Local Parallel Self-Attention (GPSA) module was introduced to enhance small-object feature representation and improve detection accuracy. To reduce computational complexity and facilitate deployment on resource-constrained edge devices, a Hybrid Adaptive Downsampling (HAD) module was designed, which effectively decreases the number of model parameters and memory consumption while accelerating inference. Furthermore, considering the low contrast and blurred details caused by turbid samples and microscopic cell structures, a Dual-path Cascaded Fusion Block (DCFNetBlock) was developed to strengthen multi-scale feature fusion and improve the capability of capturing discriminative details under low-contrast and tiny-object detection scenarios. Experimental results demonstrate that LMCF-AlgaeDet achieves Precision, F1-score, [email protected], and [email protected]:0.95 values of 88.72%, 82.81%, 80.32%, and 46.43%, respectively, outperforming the state-of-the-art YOLO26n model by 3.47, 2.23, 1.49, and 2.89 percentage points. Meanwhile, the proposed model contains only 2.633 M parameters and achieves an inference speed of 109.77 FPS, demonstrating an effective trade-off between detection accuracy and computational efficiency. Moreover, to address the scarcity of publicly available microalgae cell datasets, a new dataset containing 4994 microscopic images covering seven microalgae categories was established using the self-developed imaging system and publicly released. Overall, this study provides a low-cost, high-precision, and lightweight prototype solution for portable intelligent microalgae cell detection, offering valuable support for microalgae research and water environment monitoring applications.