Research On Leaves Recognition System Based On Texture Feature

Plant is very important to our life. Classifying these different species plants is helpful for protecting plants. Manual method of recognizing plant is low-efficiency and strongly subjective. The plant recognition based on leaf image can make up for these shortcomings and develops well.This paper researched the image processing technology applied in plant recognition, especially the texture feature extraction methods. Due to the lack of cognition about human vision and the brain, the definition and identification methods of texture have not been widely accepted, light and other natural conditions could influence the clarity of texture, and texture recognition algorithms will consume a large amount of time and memory to ensure good recognition effect. Therefore, there are still many issues need to be settled in the study of texture features.Considering some wavelet methods only focus on low-frequency sub-bands of images and some fractal dimension methods using a single exponent also cannot identify the images well, this paper proposes a novel wavelet fractal feature based approach for plant recognition. Firstly, the preprocessed leaf images are pyramid decomposed with 5/3 lifting wavelet transform and sub images are obtained. Then fractal dimensions of each sub images are calculated to be the wavelet fractal feature of leaf images. Finally back propagation artificial neural network is used to classify plant leaf images. This method is tested in Flavia leaves set and the experimental results show that the proposed method can improve the performance for plant image recognition compared with methods using only wavelet or fractal dimension, further demonstrate the importance of texture features in leaf recognition. In addition, on the basis of theoretical study, this paper also designs and implements the system of plant identification based on wavelet fractal feature.