Weed Identification using Deep Learning and Meta-heuristic Algorithms

Agriculture holds crucial economic significance and is vital for sustaining growing populations’ dietary requirements. With a substantial increase in population size, ensuring food security has emerged as a pertinent issue. The adoption of smart farming practices has become imperative to meet these objectives. This paper focuses on weed identification systems within smart farming, proposing metaheuristic-optimized Convolutional Neural Network (CNN) models. These hybrid models are empowered by algorithms like Particle Swarm Optimization (PSO), Grey Wolf Optimizer (GWO), Genetic Algorithm (GA), Salp Swarm Algorithm (SSA) and Jaya Algorithm (JAYA). These models were benchmarked against existing transfer-learning-based approaches. Notably, the GA-optimized covolutional architecture exhibited superior performance (accuracy of 96.57%) compared to pre-trained models, underscoring the effectiveness of metaheuristic optimization techniques in enhancing weed identification accuracy. This study encapsulates a comprehensive exploration of unveiling deep learning techniques to revolutionize the weed identification task.