A new ES-based feature extraction method for plant classification • Hybrid feature extraction network based plant classification system • High classification performance The proposed hybrid system includes an image-based leaf identification system consisting of three main phases which pre-processing, feature extraction, and classification. According to the working principle of feature extraction methods, a number of pre-treatment methods have been applied. Then, the z-score normalization process is applied by combining all the features obtained from these methods. Finally, the classification and testing step are carried out using the Extreme Learning Machines method. Figure A. General flow chart of recommended system Purpose: The purpose of this study is to develop a plant recognition system based on shape, texture and color features using leaf images. The proposed hybrid system aims to extract the distinctive features of the leaf. Theory and Methods: In this study, a new approach based on the edge points in the boundary curves of the leaf is proposed. This approach called edge step (ES) examines the differences between the boundary curves and geometric shapes of the leaf based on the angle, center-edge length, and edge distance characteristics. In addition, Shearlet Transform based feature extraction method was used to extract the invariant textural properties of the leaves. In addition to these methods, Gray-Level Co-Occurrence Matrix (GLCM) method and Color features were used to reveal differences between leaf vein tissues, which is an important distinguishing factor in plant classification. By combining the features obtained from all these methods, a deep and hybridbased feature extraction system has been developed for the leaves. Finally, the features obtained from all these methods were tested performances as individually and hybrid performances by using the Extreme Learning Machines (ELM) classifier method. Results: The proposed hybrid system has been tested using four leaf-based plant data sets such as Flavia, Swedish, ICL, and Foliage. According to these experimental results, the calculated accuracy values for Flavia, ICL, Swedish and Foliage datasets were 98.31%, 93.71%, 99.46%, and 96.62%, respectively. The results demonstrate that the proposed hybrid system was more successful when compared to the other studies based on shape, texture, and color features. Conclusion: In this study, a hybrid system has been proposed to reveal the distinctive features based shape and texture of the leaf. The deep-hybrid system which is developed for the classification of leaf-based plant species is practical and applicable as well as having higher performances than previous studies.
