Abstract
Crop and weed classification is crucial for weed management especially in farming industry. The objective of this study is to explore ResNet50, Convolutional Neural Network (CNN), and VGG16 architectures in order to assess how well they classify weed from crops. In order to improve performance even more, model predictions were combined using ensemble approaches such as majority voting, stacking, and averaging. The aim to assess the models generalizability, training stability, and classification accuracy. The choice of models for intricate picture classification applications is guided by this research, which advances our understanding of how ensemble techniques and model architecture affect performance. It is evident that ResNet50 and VGG16 performance is better than the baseline CNN model regarding the accuracy of training and validation, with low validation data loss and high accuracy rates. These results demonstrate the benefit of employing deeper architectures, including VGG16 and ResNet50, for enhanced feature extraction and representation in challenging picture classification tasks such as crop and weed classification. In addition to showing how ensemble techniques might improve model performance, this paper offers a thorough investigation of CNN-based architectures for image categorization. Our results establish the groundwork for future research into enhancing deep learning models for image classification tasks by providing insights into the selection and combination of architectures for high-stakes applications.