Research On Automatic Recognition Of Red Tide Algae Image Captured By Flow Cytometry Based On Linear Discriminant Analysis

Red Tide is one of the world’s three major offshore pollution problems. Its great damage to human society has caused serious concerns worldwide and many researches have been conducted in red tides. In recent years, the reports on red tides document an increase in the frequency, magnitude and geographic extent in China offshore areas, which cause billions of economic losses every year and threaten human life to death, have attracted great concerns by government, scientists and public alike. The government has invested a lot of manpower and resource on the study of the mechanism of red tide, as well as the early warning and prevention. Among all the measures to forecast and prevent red tide, a rapid identification of algae species is a key step.This thesis focuses on the recognition of red tide algae image captured by flow cytometry. Due to the problem of the collected images, such as image blurring and edge breaks, traditional approaches are difficult to guarantee the recognition quality. This paper builds a flow cytometry images automatic analysis and recognition system considering the characteristic of the images. The main work of this paper includes:1. The flow cytometry images were preprocessed to enhance the algae edge. Then an object segmentation and extraction method combining LOG operator with Otsu’s method was used to segment the red tide algae images. Compared with some traditional segmentation technology, this method can extract a more complete target algae and its contour.2. According to segmentation results,11shape features and3texture features were extracted based on the differences of the characteristics of algae species. Then verify the translation, rotation and scaling invariance of these features to determine the appropriate features.3. In-depth study of linear discriminant analysis to construct a red tide algae classification model. The shape and texture features were used for classification.4. Proposing a method to separate the adhesion cells of Skeletonema costatum by cyclic erosion, and then counting the number of cells of the algal. Applying this method on the images obtained by the classification model can get an accuracy counting rate to94.21%.This thesis constructs a red tide algae classification model based on linear discriminant analysis algorithm, which use the shape and texture features as training feature to classify the ten species of red tide algae captured by the flow cytometry. Studies show that shape features alone only bring out an average recognition rate of78.37%, while the combination of the shape and texture features can reach an average recognition rate of93.16%.