GIFF-AlgaeDet: An effective and lightweight deep learning method based on Global Information and Feature Fusion for microalgae detection

The identification and detection of microalgae are essential prerequisites for the development and utilization of microalgal resources. Traditional methods for the identification and detection of microalgae face the challenges of poor accuracy and time-consuming labor. Here is a method for microalgae identification and detection proposed in this paper, which utilizes Global Information and Feature Fusion (GIFF). Initially, to address the issue of low accuracy, the Coordinate Attention Group Shuffle Convolution (CAGS) is incorporated into the method to enhance the feature extraction capability. Furthermore, to address the issue of time-consuming labor, two small object detection heads for microalgae detection have been designed to effectively improve the training and detection speed. Additionally, the SCYLLA-IoU (SIoU) algorithm is employed to address the issue of unstable model convergence. To assess the efficacy of the method employed in this study, a dataset was intentionally created for the purpose of detecting microalgae. The experimental results indicate that, under the same experimental conditions, the proposed method has achieved significant improvements in terms of average precision, mAP@50, and mAP@95. Compared to the original method, it has increased by 3.1 %, 2 %, and 9.8 %, respectively. Moreover, this algorithm obtains a great improvement in detection speed and lightness, with a 29 % reduction in parameters and a single image detection time of 0.0219 s, which is significantly less than baseline. Location of the dataset and code: https://github.com/DjtuResearch/Microalgae_detection . • Constructed a high-quality microalgae dataset for comprehensive analysis • Proposed an effective, lightweight microalgae detection method (GIFF) • GIFF achieved 93.7 % accuracy, 95.2 % [email protected], and 0.0219 s per image.