Digital Image Processing Methods For Marine Plankton Recognition

Marine ecosystem is a complicated and changeable major system in ocean, and along with the rapid development of marine economy, changes in the marine environment systems has had a significant impact on the global biological activity and marine resource. Marine plankton is the main links in ocean web structure for marine creatures. Variations in the quantity and spatial distribution of plankton played an important role in the entire marine ecosystem and global climate change. However, the recognition and identification of plankton is a labor intensive work for marine environment monitoring and ecological study, because of the huge quantity of species with the different shapes. Therefore, the automatical identification and classification of plankton could open up a new opportunitiy in biological study.Existing methods for plankton image identification are mainly concerned with cultured algal species under controlled laboratory conditions, and there are limited works dealing with field samples which contains phytoplankton, zooplankton and suspended particles. Moreover, reports on the identification of zooplankton images are still an open area in plankton image analysis. This paper developed the techniques for image acquisition, image processing, information extraction and automatic identification of marine plankton. We also designed an online monitoring system and proposed a methodology for marine plankton image acquisition and classification in real time.Since quality of the microscopy images of living plankton are poor and there are a lot of noises in the image, in this paper, we presented a novel method based on Stationary Wavelet Transform for image denoising and enhancement. The noise suppression method is based on the properties of wavelet transforms which consists of multiresolution, multiscale, locality, time domain localization and frequency domain localization. At the same time, a combined algorithm of auto-threshold intensity segmentation, mathematical morphology operations and Canny edge detector are applied to accomplish the segmentation of plankton images.At the stage of automatic identification, the feature set consisting of global descriptions based on contour and texture information were extracted. The Principal Component Analysis (PCA) was used for feature space dimension reduction and two classification models, that one is based on Support Vector Machine (SVM) and another is based on Artificial Neural Network (ANN), were implemented. A number of experiments were performed and the experimental results showed that both of the two classification models had reasonable performance. In most of our experiments, SVM method performed relatively better than ANN classifier. The results also indicated that eliminating noise components from features can lead to a significant increase in performance in this study.To achieve the goal of capture the plankton image in real time, an online microscopy imaging monitoring system for following water was designed and constructed. This instrument can automatically provide specimens images without human intervention. Based on the automatical image acquisition system, we also proposed a framework of a real-time plankton image processing and analysis scheme. The works presented in this paper have laid the foundation for the development of a generic automated plankton identification system for practical field application.