Feature Extraction And Recognition For Algae Image

Algae play a vital role in material cycle and energy flow of aquatic ecosystem. Micro-observation, speciation analysis and counting statistics are the most important and fundamental method of algae research. Now there are lots of methods for algae detection, However, it's mainly depended on the people who have rich experiences to detect, which is not only subjectivity, poor reproducibility but also hard to operate. In the detection statistical analysis process that followed, a lot of repetitive work which is labor-intensive and low efficiency is required. The digital microscope image method based on microscope images of algal cells and computer technique has unique advantage.According to the special characteristics of the algae, an in-depth study of feature extraction, classification and recognition was made in this paper. The main contribution of this paper was summarized as follows:1)Before the extracting feature from algae image, I handle the image with Segmentation,contour extracting,color extracting. And find a suitable method for segmentation.2)we extract feature from algae image with figure, color and texture. After that, I lay out the data and compare each.3)And then, filter the invalidation feature and redundancy feature from the data.4)At last, study deeply in back propagation neural network, and improved BP neural network was proposed for getting over the disadvantages: import the momentum and automatically adjust the learning rate which can let the network avoid the local least data or the network would be collapse.This paper had taken classification experiments on five Chlorophytas: Chlorococcaceae, Micrasterias, Closterium, Arthrodesmus, Actinastrum, the recognition was above 83.3%, and three Cyanophytas: Merismopediodea, Chroococcaceae, Nostocales, the recognition was above 94.3%, and it was better than that of K-NN and SVM. The results show that the BP neural network architecture was simple and recognition rate was relatively high.