<p>In the past decade, numerous deep learning-based object detection approaches have been released and obtained desirable performance when executing in favorable weather conditions. Nevertheless, these detectors attain incomplete results of detecting objects in the rain, a frequently occurring weather condition, due to a drop in visibility and lack of crucial features for depicting objects. In this work, to bridge this gap, we publish a novel approach for elevating the efficiency of object classification and localization impaired by rain, termed UFA-Net. Our proposed approach achieves attractive object detection results via absorption sharp features at diverse scales generated from rainy images. To this end, the UFA-Net consists of three subnetworks, namely, a feature restitution (FR) subnetwork, an unsupervised learning (UL) subnetwork, and an object detection (OD) subnetwork, where, the FR subnetwork provides robust information to the OD subnetwork through the UL subnetwork. In our architecture, the FR and UL subnetworks are solely activated during training progress, and the OD subnetwork is accountable for estimating the object localization and its label in rainy weather conditions. The exhaustive experimental results substantiate that our UFA-Net achieves the highest mean average precision (mAP) scores on published light, medium, heavy, synthetic, and natural rainy image sets, respectively, surpassing compared object detection models and combinations of detectors and rain removal methods.</p>