Implementation of machine learning algorithms for accurate identification and classification of weeds

Abstract The identification of weeds remains a critical challenge in modern agriculture, where invasive species compete with crops for essential resources, leading to reduced yield and quality. Advancements in computer vision and deep learning present effective prospects for automated weed detection, enhancing precision, efficiency and sustainability in agriculture. This research explores the potential of convolutional neural networks (CNNs) for weed classification, incorporating transfer learning and lightweight architectures to achieve improved efficiency. The performance of five CNN models (VGG16 [where VGG is visual geometry group], VGG19, InceptionV3, ResNet50 [where ResNet is residual networks], and MobileNet) was assessed using a dataset of rabi season weed species ( Coronopus didymus , Fumaria parviflora , Medicago denticulata , and Rumex acetosella ). Images of these weeds were captured under field conditions using a high‐resolution mobile camera. The dataset comprised 14,418 images, partitioned into training, validation, and test sets. MobileNet emerged as the best‐performing model, achieving 94% validation accuracy and 88.87% test accuracy, with strong precision, recall, and F1‐scores across all weed classes. InceptionV3 and VGG16 also showed robust performance, while ResNet50 exhibited underfitting, suggesting a need for further optimization. The key findings emphasize MobileNet's suitability for real‐time deployment, owing to its computational efficiency and high accuracy, making it well‐suited for precision herbicide applications. Future work should aim to improve model generalization under varying field conditions, addressing dataset imbalances and integrating explainable AI techniques to facilitate farmer adoption. The study contributes to advancing AI‐driven solutions in agriculture, supporting the transition toward smarter and more sustainable farming practices.