Automated Weed Detection in Agriculture Using Deep Learning Approaches: A Comparison of CNN Architectures

The foundation of human existence is agriculture. One major problem is the presence of weeds in a crop field. Conventional weed control techniques, such as hand eradication and sporadic pesticide application, are time-consuming and environmentally damaging. This paper suggests an automated weed detection system that uses deep convolutional neural networks (CNNs) and image processing-based methods to classify weeds and crops. Evaluated several CNN architectures with an emphasis on real-time agricultural applications, such as VGG16, MobileNetV2, DenseNet121, EfficientNetB0, InceptionV3, and ResNet50. By adding more deep layers and determining the optimal CNN architecture for the weed detection system, the accuracy is increased during the field test. The best-performing model was VGG16, which obtained a test accuracy of 98.08%, precision of 97.76%, recall of 98.50%, and AUC of 98.18%.