Abstract-In this paper, a new kind of capsule endoscopy with through-body radar is utilized for the first time. Finite difference time domain (FDTD) method is used to establish an electromagnetic simulation model of stomach. A technique based on the combination of improved back-projection (BP) algorithm and support vector machine (SVM) is proposed to solve the problems of rapidly recognizing tumor shapes in the stomach. In this technique, imaging data can be obtained using the improved BP algorithm and are classified by the SVM. The algorithm must consider the influence of various tissues in the human body: the attenuation of the signal strength of electromagnetic waves, the decrease in speed and the refraction due to the different permittivity between the different organs of the body. These factors will eventually lead to image offset, and even generate a virtual image. It is effective to refrain the displacement of image with modifying the time element of the imaging algorithm by iteration. Simulation results based on data from the model verify its feasibility and validity. Results further demonstrate that the resolution is extremely high. Tumor shapes, which have different sizes, positions, and quantities, can be reconstructed using this approach. When the data are contaminated by noises, the tumor shape in the stomach can still be suitably predicted, which demonstrates the robustness of the method. Finally, classification accuracy analysis for different sampling distances and sampling intervals shows that the effects of changing the distance and intervals on shape recognition are limited. The classification accuracy can also be improved by decreasing the sampling intervals or increasing the sampling distance.