YOLO performance analysis for real-time detection of soybean pests

Abstract

In this work, we evaluated the You Only Look Once (YOLO) architecture for real-time detection of soybean pests. We collected images of the soybean plantation in different days, locations and weather conditions, between the phenological stages R1 to R6, which have a high occurrence of insect pests in soybean fields. We employed a 5-fold cross-validation paired with four metrics to evaluate the classification performance and three metrics to evaluate the detection performance. Experimental results showed that YOLOv3 architecture trained with a batch size of 32 leads to higher classification and detection rates compared to batch sizes of 4 and 16. The results indicate that the evaluated architecture can support specialists and farmers in monitoring the need for pest control action in soybean fields.

Keywords

• Deep learning
• Object detection
• Precision agriculture
• Soybean pests