Internal micro-CT image enhancement method for rice seedling stems based on a pre-enhancement network and diffusion model

In order to fix the problem of low contrast and lost detail in micro-computed tomography (micro-CT) images of rice seedling stems that have been embedded in paraffin, we propose a method for improving the images which combines a diffusion model with a pre-enhancement network. First, the triple wavelet attention (TWA) module and the selective kernel feature fusion (SKFF) module are integrated to create the multi-scale residual block (MSRB), which is the main building block of the pre-enhancement network. The overall clarity of the images is improved by this integration, which also successfully broadens the receptive field. Then, the variational information bottleneck (VIB) block is included to further eliminate noise information from the denoising diffusion probabilistic model (DDPM) and enhance its noise prediction network. Finally, the enhanced DDPM model then uses the output characteristics of the pre-enhancement network as conditional inputs to produce CT images of rice seedling stems that have a better sense of hierarchy. According to experimental data, the suggested approach outperforms the comparison methods in terms of several objective assessment metrics. Furthermore, the computational efficiency of this approach is 7.65 times greater than that of the conventional DDPM, providing robust technical support for subsequent complex computer vision applications.