Researches On Classification Of Plant Leaves Based On Images

Classification and identification of plant leaves has a great significance for accurate and efficient identification of plant species, seed identification, identification of plant leaf diseases, the genetic relationship between the study of plant and clarify the evolution.And recognition of plants based on plant leaves is of important aid for biological and ecological sciences.In recent years, based on plant leaf image classification continue to achieve progress, including features selection performance of the algorithm and classifier design of plant leaves, and has made some achievements. Overall, however, in order to improve work efficiency, reduce the labor intensity in the future, The accuracy of image-based plant leaves’automatic recognition still has room for improvement.A study for classification and identification of plant leaves in this paper. First, identification and classification and identification method based on plant leaves plant machines were reviewed,and the meaning of the classification of the plant leaves and a number of existing methods was introduced; Then, describing the basic image processing method related to the classification and identification of plant leaves; and introducing the leaf classification method used in the pre-processing and segmentation steps concisely.A leaf skeleton extraction method is proposed to help extract blade length and number of bifurcation shape features from the plant leaf images.This method introducts a significant indicator of the area of significant indicators and highlight the length of the part of the skeleton based on the existing skeleton algorithm. Using these two indicators on the skeleton of each branch to filter to remove the skeleton branch where redundant due to the small area at the edges of the projections or recesses are formed, overcoming the existing skeleton method for the edge of the regionsensitive to noise, resulting skeleton branch too much and can not efficiently reflect the regional body shape in a certain extent. Experiments show that the skeleton extraction method can effectively remove redundant skeleton branches, which gives better reflect the skeleton of the leaf area characteristics, and lays a good foundation for the further extraction of the characteristics of the blade shape, blade length, average leaf width and number of bifurcation. Finally, the paper presents a classification algorithm based on the shape and texture features, including geometric features by extracting the shape and texture features, including geometric features, Hu invariant moments, texture depicting sub and fractal dimension to form a feature vector. BP feedforward neural network is used to complete the classification, achieving higher classification rates than the existing methods in the experimental set of images. In addition, the experiments on different structure of BP network settings are carried out, and the experimental results are discussed.