Circular Characteristics Description: A Method Of Plant Leaf Image Recognition

The plant is one of the main forms of life,and its species has reached about 400000 kinds.The classification and identification of the plant is of great significance in the conservation of biodiversity,ecological agriculture and biological safety.Different types of plants generally have different leaf shapes,so the shape characteristics of the leaves play an important role in the classification of plants.However,there is a large intra-class difference and a small inter-class difference between the leaves,shapes of different leaves are very similar,so only the shape feature of leaves as the basis of recognition of plants is not enough.This paper first introduces the current popular leaf image recognition algorithms,the advantages and disadvantages of these algorithms are reviewed,the existing algorithms are classified,the most advanced recognition algorithms are analyzed and compared,and then focus on the leaf image recognition algorithm is proposed in this paper.Then,the paper focuses on the plant leaf image recognition algorithm proposed in this paper:(1)A method for image description and recognition of plant leaves based on circle feature is proposed,the paper puts forward central angle characteristics and circular integral characteristics to describe leaf shape features,and the circular gray statistical characteristics to describe the gray features of leaves,including the circular grayscale average characteristics and the circular grayscale variance characteristics.The invariance of translation,rotation and scaling are analyzed.(2)A multi-scale selection scheme is proposed,and the shape and gray featuers of the leaf are extracted by multi-scale method,thus forming a multi-scale feature vector.Combined with the discrete Fourier transform,the problem of dependence on contour starting point is eliminated.(3)The circle feature description method is tested on the Flavia leaf test set and the Swedish leaf test set,the precision / recall curve method,the mean average precision and the classification rate are respectively used to evaluate the algorithm in this paper.By adjusting the parameters in the experiment to determine the optimal number of scales and the optimal weights,the effectiveness and superiority of the proposed algorithm is proved by comparing with other leaf recognition algorithms.