Application of Object Detection Algorithm in Identification of Rice Weevils and Maize Weevils

Deep learning based models have had great success in object detection, but the state of the art models have not yet been widely applied to the identification of the stored-grain pests. We apply for the first time an object detection model to identify rice weevils and maize weevils, which have always been a challenge in the field of the research of stored-grain pests because of their very similar appearance. To conduct our initial study, we created a pre-training dataset of 4000 images and a object detection dataset of 1600 images. In the experiments, we used Faster R-CNN and R-FCN as object detectors and used VGG16, ResNet101 and Inception-ResNet-v2 as feature extractors. In detail, we pre-trained the object detection models on our pre-training dataset, and fine tuned with our object detection dataset. Finally, we demonstrate that the final object detection model outperforms our baseline and shows a nice detection effect with a high accuracy. It is worth noting that our research will have a revelatory influence on stored-grain pest control and grain storage.