Marine zooplankton has important ecological and economic value. The observation and automatic image recognition technology of marine zooplankton is an important mean to acquire data such as species, quantity, spatial distribution and behavioral postures of zooplankton, and is an important support for marine scientific research. Digital holography has an innate advantage of refocusing and reconstruction, which is suitable for deep learning and living zooplankton recognition. In this study, a large number of holographic images was trained by using the improved YOLOv2 model, and after test, the study achieved satisfactory results: the models trained by the images with sharpness assessment score of 0.6 or higher, have precision rate above 94% and a recall rate above 88%. This study mainly discusses: (1) the detection method of moving targets to acquire the images of moving zooplankton; (2) the two factors that affect the holographic images recognition results, mean (pixel mean of images) subtraction operation and image sharpness, and the no-reference sharpness assessment based on structural similarity for holographic images; (3) the relationship between sharpness assessment index or mean subtraction and the recognition results.

The motion behaviors of copepods has important scientific research value and there is very little research on recognition of their motion behaviors simultaneously. Recognition of the basic motion behaviors of copepods using deep learning methods can greatly reduce the time cost of distinguishing and statistics, as well as achieve the purpose of improving efficiency. Based on the characteristics of motion of copepods that bring challenges to the extraction of motion fragments from raw video and the establishment of data set, such as instantaneous moving, static status most time, small-scale and high-frequency, this paper propose an improved Camshift algorithm for detection of moving targets to overcome these challenges and establish the motion behaviors image acquisition system and a standard data set of motion behaviors, which provides the experience and methods of marine zooplankton behaviors database. Finally, the LRCN network that combines the advantages of CNN and LSTM is adopted to study the impacts of different factors on the model performance, such as the number of frames of sample, preprocessing operations and sample dimensions. Experimental results show that the LRCN network has excellent potential in classification of motion behaviors of copepods, when the number of frames of sample reaches 7, the precision, recall, f1-score are 0.96, 0.95, 0.95, respectively. In addition, the rise in number of frames and preprocessing has a positive effect on the recognition, the 4D samples (image sequence) is more suitable for the LRCN model than 3D samples (trajectory image).