Abstract
Object detection in remote sensing, especially in aerial images, remains achallenging problem due to low image resolution, complex backgrounds, andvariation of scale and angles of objects in images. In current implementations,multi-scale based and angle-based networks have been proposed and generatepromising results with aerial image detection. In this paper, we propose anovel loss function, called Salience Biased Loss (SBL), for deep neuralnetworks, which uses salience information of the input image to achieveimproved performance for object detection. Our novel loss function treatstraining examples differently based on input complexity in order to avoid theover-contribution of easy cases in the training process. In our experiments,RetinaNet was trained with SBL to generate an one-stage detector,SBL-RetinaNet. SBL-RetinaNet is applied to the largest existing public aerialimage dataset, DOTA. Experimental results show our proposed loss function withthe RetinaNet architecture outperformed other state-of-art object detectionmodels by at least 4.31 mAP, and RetinaNet by 2.26 mAP with the same inferencespeed of RetinaNet.