Reproduce “Integrated Image Reconstruction and Target Recognition based on Deep Learning Technique” on OpenResearch

About this paper

Cien Zhang, Jiaming Zhang, Jiajun He, Okan Yurduseven

Att-ClassiGAN, a deep learning model, enhances the ClassiGAN architecture by integrating attention gate mechanisms to perform simultaneous microwave image reconstruction and target classification from backscattered signals. It achieves an NMSE of 0.018 and SSIM of 0.983 for reconstruction, 0.980 classification F1-score, and reduces reconstruction time to 0.059 seconds.

Key insights

• Integrating attention gate modules into the deep learning architecture enables the model to focus on relevant features, enhancing both image reconstruction quality and target classification accuracy.
• A single Generative Adversarial Network (GAN) architecture can simultaneously perform both microwave image reconstruction and target classification from backscattered signals.
• Utilizing an uncertainty-weighted loss function allows for adaptive balancing of the image reconstruction and classification tasks during training, optimizing overall performance.

Summary by alphaXiv.