A systematic review of machine learning and deep learning approaches in plant species detection

Due to global warming and adverse climatic conditions, researchers have focused on the survival of plant species. Certain plant species are nearly extinct and thus must be saved. However, this requires a comprehensive understanding of the species, which involves expertise and time. Many researchers have expressed their interest in developing automated plant species recognition techniques by analyzing various issues, such as expertise, complex venation structure, interclass and intraclass similarity, occluded leaves, problems with feature extraction of premature and dried leaves, etc. Classifiers and other machine learning and deep learning algorithms that can accurately categorize and extract data can be further developed. In this article, we presented a systematic review of effective machine learning methods and image processing and deep learning algorithms for plant species recognition that have been used by researchers. The most common research problems, recommendation techniques, methodological analysis, and datasets presented in these selected studies were examined. The review and analysis indicated that there is scope for improvement in the case of multi-feature fusion, such as concatenation of shape, size, margin, texture, venation contour, etc. The extraction of characteristics of leaves in different stages, such as seedlings, tiny, mature, and dried, can be considered for feature extraction. Hybridization of machine intelligence approaches can also be implemented for multifeatured fusion extraction which might increase the performance of the system. We also investigated evaluation metrics, challenges, trends, and the most interesting ideas for future studies.