In this article, we design and implement RF-Ray, a system that can detect whether the internal object of a package is illegally replaced using radio frequency (RF) identification tags in dynamic transport without opening the package, preventing the replacement of objects and enhancing the safety of the object in transport. RF-Ray is based on the principle that different objects have different reflective features to tag signals, and we extract the reflective features of the object from multipath signals by building a feature extraction model, which is based on a convolutional neural network. We propose solutions to all challenges encountered and conduct extensive experiments to verify system's accuracy. Our results show that RF-Ray can achieve an accuracy rate of over 97%. Compared with Echoscope, TagRay, and Tagscan, RF-Ray can guarantee high accuracy to identify the internal object in the package in dynamic scenes. In addition, we conduct an experimental analysis of factors that affect the performance of the system, and experiments demonstrate that RF-Ray has a good performance.