Research On Automatic Identification Of Micro Green Algae

Algae, usually living in water, is a primary plant-like creature of the kingdomProtista. On Earth90%of photosynthesis is completed by algae, which can be dividedinto11categories. Chlorophyta is a vitally important category in algae, on which theresearch is closely related to the green food, green energy, biopharmaceuticals,environmental protection and other fields. But the traditional identification of algae,often done under a microscope, requires the operators highly mastering algae forms,and the work intensity is so hard that can easily caused the subjective error. In thispaper, we use the digital image processing technique to study on the automaticextraction and identification of Chlorella micro-images, the major tasks are asfollows:Firstly, we introduced the significance of the research and the research status athome and abroad. Traditional green algae classification and the deficiencies aredescribed, and then we introduced the problem about the current systems based ondigital image processing in the algae identification.Secondly, we detail the method how we get the green algae target in themicroscopic image. The procedure is graying the image in B-channel, and thensegmenting the image by Otsu method, obtained the binary image of the area ofinterest is more prominent. Add an enhancement by using morphology operations,after which use the contour tracking in combination with filtering algorithm to obtainthe target area. According to the main contours, figure out the minimumcircumscribed circle and the main spindle, rotate the image about the spindle, figureout the minimum enclosing rectangle.Thirdly, propose the classification method based on the divided concentriccircles. Divide Image of minimum radius of circumscribed circle is divided into10concentric feature extraction circles, extract Histogram feature on the variousconcentric circles, target feature of axial rectangular, gray level co-occurrence matrixcharacteristics of concentric circles of3rd and4th division, three eigenvectors representing the color, structure and texture features of the target. Then, sort all thedatabase images according to the distance and figure out the final result by the KNN.The innovation of this paper is using gray level co-occurrence matrixcharacteristics of concentric circles of3rd and4th division and combing imageretrieval with classifiers. Evaluation of this paper contains the classification accuracyusually used in classifier, sorting precision and recall precision usually used in imageretrieval. Experiment in the gallery there are over186images in60species. Theaverage classification accuracy of this method is94%(full100%), average sortingaccuracy rate is5.46%(full6%), average recall accuracy is92.4%(full100%), thismethod achieves better result compared with other similar methods.