Model Accuracy Analysis: Comparing Weed Detection in Soybean Crops with EfficientNet-B0, B1, and B2

Abstract

Comparison of EfficientNet models B0, B1, and B2 for soybean weed detection. This study examines how well these models distinguish weeds from soybeans. Precision, recall, and F1 score metrics evaluate each model through extensive testing and experiments. The method trains these models on datasets with images of soybean fields overrun by various weeds. Results show subtle differences in model accuracy, showing what they're good at and what they can't do to find soybean weeds. EfficientNet-B2 detects and classifies weeds better than B1 and B0 in soybean fields. It could improve weed management systems' accuracy and reliability. This comparison helps us choose the best weed-finding model to improve soybean farming and prevent crop yield losses. Training accuracy measurements on EfficientNet models B0 = 100%, B1 = 100%, and B2 = 100%.