Early-stage weed detection in farmland imagery plays a critical role in precision agriculture, helping minimize herbicide usage and maximize crop productivity. CNNs have shown potential in this domain, but their performance heavily depends on optimal configuration of hyperparameters and feature selection. However, conventional CNN models often struggle with overfitting, high computational cost, and suboptimal accuracy due to manual or heuristic tuning of parameters. These limitations hinder their reliability in complex field environments where weed and crop appearances vary significantly. To address these issues, we propose an Integrated Particle Swarm Optimization with Convolutional Neural Network (I-PSO-CNN) framework. This method leverages the global search capability of PSO to autonomously fine-tune CNN hyperparameters, including learning rate, number of filters, and convolutional kernel size, thereby improving feature learning and generalization. The optimized model is applied for real-time detection and classification of early-stage weeds in Unmanned Aerial Vehicles (UAV)-acquired farmland imagery. This allows for accurate weed mapping and facilitates targeted herbicide application. Experimental results demonstrate that the I-PSO-CNN model outperforms traditional CNNs in terms of detection accuracy, training efficiency, and robustness to image variations, making it suitable for large-scale agricultural deployments.

Keywords :

Weed Detection, Precision Agriculture, Convolutional Neural Network, Particle Swarm Optimization, UAV Imagery, Smart Farming.