Incorporating the Completeness and Difficulty of Proposals Into Weakly Supervised Object Detection in Remote Sensing Images

Weakly supervised object detection (WSOD) in remote sensing images (RSI) only require image-level labels to detect various objects. Most of the WSOD methods incline capture most discriminative parts rather than entire object, and number easy hard samples is imbalanced. To address first problem, a novel metric named objectness score (OS) proposed incorporated into training loss our model. The OS consisted traditional class confidence (CCS) completeness prior (OCPS). CCS can provide probability that proposal belongs certain class, OCPS quantify covers object. Therefore, which cover with high confidences will be assigned large weight through OS. handle second difficulty evaluation (DES) also loss. DES calculated by using entropy vector each used how difficult identified correctly, consequently, DES. ablation experiments on two RSI datasets verify effectiveness comprehensive quantitative subjective evaluations demonstrate method inclines accurately, surpasses seven state-of-the-art methods.