Enhanced Recurrent Neural Network With C-Efo-Based Feature Selection for Plant Leaf Classification

Abstract

Plants are an important factor in human life and other living things around the world. Plants are recognized as important influencers of changes in natural cycles. It is an important producer that sustains human life, as it is known to be the only organism capable of converting light energy obtained from the sun into food energy for humans and other organisms. Animals cannot produce food because they depend directly and indirectly on plants for food energy. Automated plant recognition seeks more attention in computer vision and machine learning. A lot of research has been done to solve the problems related to plant classification. The knowledge and ability to distinguish various medicinal plants was locked in by early people before the development of computer systems and digital cameras. The new plant leaf classification was developed in the first stage using improved segmentation techniques and optimal feature selection. Experimental images were collected from the Swedish leaf dataset and subjected to a preprocessing step. The preprocessed image is obtained through grayscale conversion, median filtering, and histogram equalization. Therefore, an optimized UNet model is used to obtain key regions of leaves to improve accuracy. Features of shape, texture and color were obtained. Since they contain the longest length of the resulting features, the best features are chosen to reduce training time and dimensionality reduction. These optimal characteristics are achieved through a modified hybrid algorithm called C-EFO (Crow Search Electric Fish Optimization), where the traditional EFO (Electric Fish Optimization) is combined with the CSO (Crow Search Optimization) algorithm. Once the best features were obtained, the newly developed E-RNN deep learning model was used for classification, where the hyperparameters were best fitted using the C-EFO algorithm. Finally, the experimental results are validated and the proposed model achieves better performance metrics. Experiments show that the proposed C-EFO method outperforms traditional methods in terms of accuracy.