Plant leaf recognition algorithm using ant colony-based feature extraction technique

Plant recognition as a substantial subject of biology has occupied the minds of many botanists throughout the world to concentrate their efforts on the identification of unknown plant species with the aim of protection and other purposes. As a troublesome and gradual process, traditional methods of taxonomy of plants impede a high rate of performance for the taxonomist in this field. In the modern-day, improvements in the fields of artificial intelligence and soft computing have led to the field of automatic plant recognition being considered as a challenging topic due to the various uses of plants in medicine, food and industry. Although many studies have been undertaken to seek out a method that can be applied for the classification of numerous plants, there is still a lack of a highly efficient system for the recognition of a wide range of different plants. The aim of this research is to contribute to the measurement of physiological dimensions of plant leaves by the proposed Auto-Measure algorithm to operate in an automatical manner which inherently requires an improvement in automatic feature extraction. Moreover, the ant colony optimisation technique be applied as an expert algorithm to make a decision for the selection of optimal features in order to enhance the performance of a classifier for recognition of diverse species of plants. To do this, at first, based on the proposed algorithm,the physiological dimensions of leaves are automatically measured and with regard to these parameters, specified features such as shape, morph, texture and colour are extracted from the image of the plant leaf through image processing to create a reserved feature database to be used for different species of plants. Then, based on the characteristics of each species, decision making is done by means of ant colony optimisation as a search algorithm to return the optimal subset of features regarding the related species. Finally, the selected features are employed by a multi-class support vector machine to classify the species. The proposed method was applied to different kinds of plant and herb species for testing the system and it was found from the experimental results that the system, by eliminating redundant features, not only optimised the number of features in the subset, but also had a remarkably positive impact on the performance of the classifier in a way that implementation of the proposed method on almost 2830 leaves improved the average accuracy over all the plant databases to 96.66 %. Therefore, it can be concluded that the proposed method is capable of a high rate of classification of various plant species.

Preview PDF FK 2013 34R.pdf Download (808kB) | Preview

Actions (login required)

View Item