BSoNet: Deep Learning Solution for Optimizing Image Quality of Portable Backscatter Imaging Systems

TL;DR

This work tackles poor image quality in portable backscatter imaging (PBI) caused by limited backscattered photons. It introduces BSoNet, a framework that combines Resolution Adaptive Network (RANet) and Backscatter Optimizing Transformer (BSformer), empowered by Noise2Void-based self-supervised training to operate without clean labels. The approach yields clearer, higher-contrast images with improved target recognition, validated on PBS-140 data and capable of real-time remote deployment. The results suggest a practical path to robust, high-quality PBI imagery for security screening and industrial inspection tasks.

Abstract

Portable backscatter imaging systems (PBI) integrate an X-ray source and detector in a single unit, utilizing Compton scattering photons to rapidly acquire superficial or shallow structural information of an inspected object through single-sided imaging. The application of this technology overcomes the limitations of traditional transmission X-ray detection, offering greater flexibility and portability, making it the preferred tool for the rapid and accurate identification of potential threats in scenarios such as borders, ports, and industrial nondestructive security inspections. However, the image quality is significantly compromised due to the limited number of Compton backscattered photons. The insufficient photon counts result primarily from photon absorption in materials, the pencil-beam scanning design, and short signal sampling times. It therefore yields severe image noise and an extremely low signal-to-noise ratio, greatly reducing the accuracy and reliability of PBI systems. To address these challenges, this paper introduces BSoNet, a novel deep learning-based approach specifically designed to optimize the image quality of PBI systems. The approach significantly enhances image clarity, recognition, and contrast while meeting practical application requirements. It transforms PBI systems into more effective and reliable inspection tools, contributing significantly to strengthening security protection.

Figures (8)

• Figure 1: Schematic diagram of the portable backscatter imaging system, demonstrating the Compton scattering effect caused by the incident photon interacts with the electrons of the target material, resulting in forward scattered photons (at an angle of ${{\theta }_{3}}$) and backscattered photons (at angles of ${{\theta }_{1}}$ and ${{\theta }_{2}}$). The backscattered photons are received by the device detector.
• Figure 2: Schematic diagram of the structure and some components of the PBS-140 portable backscatter imaging system. (a) Front view of PBS-140. (b) Top view of PBS-140.
• Figure 3: BSoNet, a deep learning-based imaging quality optimization framework for PBI and its key components. RANet: Resolution Adaptive Network. BSformer: Backscatter Optimization Transformer. FLN: Feature Learning Network structure within BSformer. FFN: Feature Fusion Network structure within BSformer. CA: Channel Attention module within RANet. DCR: Dense Connection Block network structure within FLN. AAP: Adaptive Average Pooling layer. AMP: Adaptive Max Pooling layer.
• Figure 4: Visual optimization effects of BSoNet and other optimization methods on several representative cases (including quantitative results of local contrast). Each set of images illustrates a specific case. Our method is highlighted with a red box, and the relative position of each method is marked in detail in group (a) to facilitate observation and comparison of the effects of different processing techniques.
• Figure 5: A simulated scenario of hiding contraband in car seats during border inspection, where dog teeth and cattle horns serve as substitutes for smuggled animal products, and white sugar is used to simulate drugs.