Performance of deep learning models for classifying and detecting common weeds in corn and soybean production systems

Knowing precise location and having accurate information about weed species is a prerequisite for developing an effective site-specific management (SSWM) system. Due to the effectiveness of deep learning techniques vision-based tasks such as image classification object detection, its use discriminating between weeds crops gaining acceptance among agricultural research community. However, limited studies have used identifying multiple in single most not compared based detection by using common, annotated imagery dataset early season under field conditions. This study addresses gap evaluating comparative performance three different pre-trained models classifying also assesses model locating species. The were trained on two commonly frameworks i.e., Keras PyTorch, assess any differential due choice framework. An comprising RGB images four, weeds, found corn soybean production system Midwest US, namely, cocklebur (Xanthium strumarium), foxtail (Setaria viridis), redroot pigweed (Amaranthus retroflexus), giant ragweed (Ambrosia trifida) was this study. VGG16, ResNet50, InceptionV3 classification. model, You Only Look Once (YOLOv3) library, locate identify within image. assessed testing accuracy F1-score metrics. Average precision (AP) mean average (mAP) model. best performing VGG16 with 98.90% 99%. Faster training times higher accuracies observed PyTorch. helped AP scores 43.28%, 26.30%, 89.89%, 57.80% cocklebur, foxtail, pigweed, respectively overall mAP score 54.3%. results suggest that conditions, YOLOv3 are promising respectively, given sufficient data available. Additionally, unlike classification, localization capabilities desirable SSWM.