Object detection approaches have made significant advances in favorable conditions of weather. Nevertheless, this success is difficult to replicate with rainy images due to reduced visibility. To address this challenge, we develop a novel framework that is improved from YOLOv7 to enhance the accuracy of detecting objects covered by rain. The proposed feature recovery (FR) module is designed to remove rain for generating clear features. We plug the FR module into the head part of the base detector to raise the exploiting feature capabilities of the model, aiming to gather useful information for locating various objects. In addition, the feature fusion method is applied in our method to enrich knowledge at different scales for detecting instances of diverse sizes. Experimental results display that our method obtained the highest object detection performance of 72.24%, 57.83%, and 76.61% on three different published rainy datasets: iRain, RID, and rRain, respectively, and exceeded state-of-the-art (SOTA) object detection models.