Accurate quantitative plankton observation is significant for biogeochemistry and environmental monitoring. However, current observation equipment is mostly shipborne, and there is a lack of long-term, large-scale, and low-cost methods for plankton observation. This paper proposes a solution to investigate plankton using a Seascan holographic camera equipped with a “Petrel-II” underwater glider for a longer time sequence and at a larger scale. Aiming at the new challenges of low efficiency and low accuracy of holographic image processing after integrating holographic imaging systems and underwater gliders, a novel plankton data analysis method applicable to Digital Holographic Underwater Gliders (DHUG) is proposed. The algorithm has the following features: (1) high efficiency: the algorithm breaks the traditional hologram information extraction order, focusing only on the key regions in the hologram and minimizing the redundant computation; (2) high accuracy: applying the Sobel variance algorithm to the plankton in the hologram to focus the plane extraction significantly improves the focus accuracy; and (3) high degree of automation: by integrating a convolutional neural network, the algorithm achieves a fully automated analysis of the observed data. A sea test in the South China Sea verified that the proposed algorithm could greatly improve the problems of severe plankton segmentation and the low focusing accuracy of traditional information extraction algorithms. It also proved that the DHUG plankton survey has great potential.

The marine ecosystem is one of the most extensive and abundant ecosystems on Earth. Marine plankton is an important component, and its abundance, number of species, and dominant species are regarded as important monitoring indicators. Aiming at the problems of low accuracy and high complexity in identifying plankton based on convolutional neural networks, this study proposes a lightweight identification algorithm for plankton images based on the improved MobileNetV2. Firstly, the network layer structure is extracted by redesigning features to balance the depth and width of the network to reduce the model parameters; secondly, the lightweight coordinate attention (CA) mechanism is introduced to strengthen the attention and extraction ability of key areas; in addition, the structure of the network classifier is optimized to improve the utilization efficiency of the model parameters. The results show that the model achieves a 95.46% accuracy and 94.48% recall in 12 kinds of images. Compared with the initial MobileNetV2, the parameters and calculation amount are reduced by 72.47% and 52.09%, respectively, and the reasoning time for a single image is 6.15 ms. The model realizes the accurate identification of plankton in situ under the premise of ensuring it is lightweight. Combining time information and depth data, it is of great significance for marine ecological environment monitoring and prediction to obtain the abundance of various plankton.