Detection of small objects based on feature fusion

In order to solve the problem of poor detect effectiveness of small target objects in the process of algorithm, a feature fusion method for Faster R-CNN has been proposed. This method fully fuses the deep and shallow feature information, which well improves the detection model for small objects. Meanwhile, in order to better detect small objects, oversampling is used to preprocess the data, and the corresponding hyperparameter values of the Faster R-CNN model are adjusted. From the experimental results, it is easy to see that the detection accuracy is improved by 7.6%, and for small target objects Bottle, Plant, Cow and Boat is improved by 13.9%, 11.2%, 6.7% and 9.5%, respectively. The detection effect of this model has been substantially improved.