Artificial Intelligence-Based Site-Specific Weed Management in Cotton: A Comprehensive Review
Vaibhavkumar R. Dama
*
Department of Farm Machinery and Power Engineering, College of Agricultural Engineering and Technology, Junagadh Agricultural University, Junagadh, Gujarat, India.
T. D. Mehta
Department of Farm Machinery and Power Engineering, College of Agricultural Engineering and Technology, Junagadh Agricultural University, Junagadh, Gujarat, India.
N. B. Parmar
Department of Farm Machinery and Power Engineering, College of Agricultural Engineering and Technology, Junagadh Agricultural University, Junagadh, Gujarat, India.
B. V. Patoliya
Junagadh Agricultural University, Junagadh, Gujarat, India.
B. M. Khanpara
Department of Agricultural Engineering, College of Renewable Energy and Environmental Engineering, Sardarkrushinagar Dantiwada Agricultural University, Sardarkrushinagar, Gujarat, India.
C. S. Matholiya
Department of Farm Machinery and Power Engineering, College of Agricultural Engineering and Technology, Junagadh Agricultural University, Junagadh, Gujarat, India.
N. B. Bharad
Department of Farm Machinery and Power Engineering, College of Agricultural Engineering & Technology, Junagadh Agricultural University, Junagadh -362001 (Gujarat), India.
R. V. Bharadava
Department of Farm Machinery and Power Engineering, College of Agricultural Engineering & Technology, Junagadh Agricultural University, Junagadh -362001 (Gujarat), India.
*Author to whom correspondence should be addressed.
Abstract
Weed infestation is one of the major biotic factors limiting cotton (Gossypium spp.) productivity, particularly during the early growth stages, where uncontrolled weed competition can result in yield losses ranging from 30 to 80%. The conventional weed management practices, such as manual weeding, mechanical cultivation and uniform herbicide application, are labour-intensive, costly and often environmentally unsustainable. In recent years, Site-Specific Weed Management (SSWM) integrated with Artificial Intelligence (AI) has emerged as an advanced precision agriculture approach for real-time weed detection and targeted herbicide application.
This review comprehensively summarizes recent developments in AI-based weed detection, machine vision systems, deep learning techniques and intelligent spraying technologies in cotton production systems. Earlier image-processing methods based on colour indices, shape descriptors and thresholding techniques achieved moderate weed classification accuracies ranging from 52 to 74% and showed limited adaptability under variable field conditions. The subsequent adoption of machine learning approaches, including SVMs, ANNs and wavelet-based feature extraction techniques, significantly improved weed identification performance, with accuracies exceeding 98% under controlled conditions. More recently, deep learning models, particularly CNNs and real-time object detection frameworks such as YOLO and Faster R-CNN, have demonstrated superior crop-weed discrimination capabilities with detection accuracies above 90 to 95% under dynamic field environments.
The integration of AI-enabled weed detection systems with precision spraying technologies, including variable-rate and micro-jet sprayers, has substantially improved herbicide application efficiency and operational accuracy. Several studies have reported herbicide savings ranging from 40 to 90% through site-specific spraying while maintaining effective weed control and reducing environmental contamination. In addition, intelligent sprayer systems equipped with embedded computing platforms, sensor fusion technologies and automated control mechanisms have facilitated real-time field implementation and enhanced input-use efficiency.
Despite considerable technological advancements, several challenges continue to hinder large-scale adoption of AI-based weed management systems. These include variability in illumination conditions, morphological similarities between crops and weeds, limited availability of annotated datasets and computational limitations associated with real-time field operations. Furthermore, there is a lack of region-specific AI models designed for Indian cotton production systems and diverse agro-climatic conditions.
Overall, this review highlights major research advancements, existing limitations and future opportunities in AI-driven weed management for cotton cultivation. The integration of robust datasets, adaptive learning algorithms, edge computing, robotics and IoT-enabled autonomous spraying systems can contribute significantly to the development of sustainable, economically viable and environmentally responsible weed management strategies in cotton production systems.
Keywords: Artificial Intelligence (AI), site-specific weed management (SSWM), deep learning, convolutional neural networks, YOLO object detection, machine vision, precision agriculture, variable rate spraying, smart sprayer.