Deep Learning with Harris Hawks Optimization for Types of Weed Classification

Abstract

Deep learning, a branch within machine learning, aims to represent abstract ideas into data by placing multiple layers of processing [1]. It has been widely applied in various fields, particularly in image classification, and the best advantage of deep learning is its ability to aperient automated image classification through a well-trained neural network, which can be applied to immense array of image types [2]. The Harris Hawks optimizer is distinct as a well-known optimization algorithm based on swarm intelligence that does not request gradients [3]. It has attracted considerable interest from the researchers because of its performance, quality of results, and reliable convergence when dealing with many applications in diverse fields such as medicine, network systems, and image classification [4]. In this paper, the classification performance of different types of weeds was improved. The EfficientNetB0 model implements an architecture for weed classification. We then used the DenseNet121 model architecture as another option for the classification of weeds. In addition, a CNN (Convolutional Neural Network) model has been used to develop a custom CNN architecture tailored to the weed classification. Finally, the (Hary Hawks Optimization) HHO with CNN (CNN-HHO) model was applied to explore the integration of Hary Search Optimization with CNN for improved performance. Experiment findings reveal that our presented method outperforms and achieves 99% accuracy.