Research On Automatic Recognition System To Harmful Algae Based On Double-Spectrum

The red tide is a phenomenon that seawater changes color caused by the excess reproduce or assembling of harmful algae, bacteria and protozoan .The red tide is a serious global marine disaster causing billions of dollars'economic loss to the world. So it is very important to prevent and monitor it. However, at present, the identification and measurement mainly observed and analyzed in the lab artificially. But the artificial observation and analysis can not satisfy the prediction of the red-tide. The paper just bases on the fact, and aims at to develop an automatic system recognizing and classifying the harmful microalga, which will be very significant and make the prediction work of red ride much easier.According to the auto fluorescent nature of harmful algae, the paper developed an automatic system of recognizing and classifying harmful algae based on the double-spectrum technology. The system includes two parts, hardware and software. The hardware system includes microscope, fluorescence switch device, CCD , image card, computer. The software system includes image pre-processing, image segmentation, image feature extraction, image feature selection , establishment of harmful algae information database, recognizing algae using neural network. The software is the core of the system. The software accomplish the recognition of algae using artificial neural network based on the difference of the shape, texture, and contour between different kinds of algae.The double-spectrum technology is very important in the paper, which means fluorescent image and grey-scale image of harmful algae. It is an effective method to eliminate background yawp combining the advantages of two kinds of image. The advantage of fluorescent image is that the background is clear. The advantage of grey-scale is that the contour and texture of harmful algae is intact. Thus,in the process of image analysis, firstly we take fluorescent image and grey-scale image of harmful algae under same field of view , then we process the fluorescent image and mark the position of algae. Then combining the position information of fluorescent image with the contour and texture information of grey-scale image, we can use computer to recognise and count algae automatically.The paper did recognition experiments with 6 kinds of harmful algae including Gymnodinium Stein, Saline Algae, Noctiluca scientillans, Heterosigma akashiwo, Coscinodiscus gigas Ehrenberg and PR Algae. The recognition accuracy is up to 90%, as compared with the human recognition.