Researches On Recognition For Microscopic Image Of Chaetoceros And Counting For Multiple-Cell Of No-setae Algae Species

In recent years, the frequency of harmful red tide is becoming higher and higherin coastal waters of China. The red tides damage the marine environment severely,and impact on coastal marine economy and human health. Monitoring harmful algalblooms (HABs), water quality assessment and aquatic ecosystem research are allbased on the qualitative and quantitative research of the HABs. This thesis analyzesthe bio-morphological characteristics and microscopic image characteristics ofChaetoceros and no-setae algae species which are common in Chinese coast, andbuilds automatic recognition system of Chaetoceros and automatic counting system ofno-setae algae species based on the image processing technology. The main work ofthe article has the following3points:1. Object extraction of Chaetoceros microscopic image. The proposed method givesenough consideration to the feature of setae，the particular band pattern of darkand light, which is distinguished from noise. The model of vectorizationrepresentation of gray surface is constructed by using the knowledge of the spacevector，and the gray images both horizontally and vertically are obtained byvector mapping. The algorithm can solve the noise disturbing and reserve thesetae information. In contrast with others, the proposed method can extract moresetae information.2. Recognition for microscopic image of Chaetoceros.12moment invariant featuresand7shape features using the moment invariants and parameters of shapecharacteristic are extracted as19-dimensional feature vectors. SVM is used as aclassifier of Chaetoceros. The experiment has trained with1850images covered18species, and965images are used to verify the effect, the average recognitionrate is83.1%. 3. Counting for multiple-cell microscopic image of no-setae algae species. A methodcombined the Canny operator and Otsu algorithm of area is proposed to segmentcells in no-setae algae species microscopic image. For the presence of cell clumpsof touching or overlapping objects, detecting corner points，listing candidate splitlines and finding the best split line are used to split clumps. Finally, the image ofobject extraction is labeled to calculate the algae number by connectedcomponents. The counting results are compared with the watershed algorithm andanalyzed quantitatively.115images which cover15species, and include684cellsare applied in the experiment, the precision is88.1%, the recall rate is94.7%, theF1-measure is91.3%.The researches in this article are efficient in extracting more setae, recognizingChaetoceros and automatic counting of no-setae algae species.