Comparative Analysis of Various Deep Learning Techniques for Weed Detection in Corn Crops

Weed detection is an important process in increasing the crop yield and minimizing labor input in crop management. Automatic weed detection helps in enhancing precision agriculture. The objective of this study is to explore the application of deep learning techniques for automated weed detection. It compares the performance of four deep learning models: Standard CNN, Custom Residual CNN, VGG16, and EfficientNet B0. The models were trained with a publicly available corn dataset with 24,000 images with various image processing techniques such as normalization and data augmentation to increase the model’s resilience. EfficientNet B0 performed better when compared to other models with an accuracy of 96% proving the higher capability for crop and weed classification.