Robust quantization watermarking with perceptual JND model has made a great success for image copyright protection. Generally, either restores each color channel separately or processes the vector representation from three color channels with the traditional monochromatic model. And it cannot make full use of the high correlation among RGB channels. In this paper, we proposed a robust quaternion JND Model for color image watermarking (QuatJND). In contrast to the existing perceptual JND models, the advantage of QuatJND is that it can integrate quaternion representation domain and colorfulness simultaneously, and QuatJND incorporates the pattern guided contrast masking effect in quaternion domain. On the other hand, in order to efficiently utilize the color information, we further develop a robust quantization watermarking framework using the color properties of the quaternion DCT coefficients in QuatJND. And the quantization steps of each quaternion DCT block in the scheme are optimal. Experimental results show that our method has a good performance in term of robustness with better visual quality.

In the process of ore crushing, the accuracy of ore image segmentation directly affects the validity of particle size distribution detection. In order to solve the problems of low accuracy and high model complexity of existing segmentation methods, an efficient attention decoding U-Net (EU-Net) model is proposed in this article to demonstrate higher sensitivity towards subtle boundary. By focusing on the effective feature mapping of the ore and the slice-level hierarchy fusion, an efficient mapping hierarchy fusion (EffiMHF) module is designed to reduce the number of model parameters and redundant calculations. In order to solve the adhesion problems in ore image segmentation, we propose an adaptive complementary fusion segmentation (ACFS) method, which can segment more uncertain regions by relaxation prediction. Experiments show that the combination of EU-Net and ACFS can effectively solve the problem of adhesion in segmentation, making segmentation more accurate and efficient. Compared to the eight state-of-the-art segmentation models, EU-Net achieves 86.13%, 88.00% and 79.16% in metrics Accuracy, F1_score and IoU, respectively, which is the outperforming and has the least parameters.

Ore image segmentation plays a vital role in mineral processing, directly affecting the accuracy of crushing quality assessment and particle size analysis. However, the accuracy of traditional segmentation techniques is severely compromised by three inherent challenges in ore images: the wide size range of particles, significant color variations within individual ores, and indistinct boundaries between agglomerated particles, leading to imprecise segmentation results. To overcome these challenges, this study introduces LAES-UNet, an integrated network based on EfficientSAM, which combines a pre-trained EfficientSAM encoder with a multi-level decoder. The proposed architecture incorporates three tailored modules: the local-global hierarchical interaction for multi-scale feature enhancement, the adaptive spatial feature refinement for adaptive weighting across color-variable regions, and the edge focusing module for explicit edge and fine-detail perception. Experiments were conducted on a self-built conveyor belt ore image dataset containing 148 manually annotated images, which were cropped into 296 non-overlapping samples, as well as on a public mineral image benchmark. The results show that LAES-UNet achieves the best overall segmentation performance among the compared methods, with up to 2.8% higher intersection over union (IoU) and consistently improved boundary delineation. Furthermore, a kernel density estimation-based particle size distribution fitting method verifies the practical value of the segmentation results in quantitative ore particle size analysis. Overall, LAES-UNet delivers a generalizable solution for automated, high-precision particle size measurement within intelligent mineral processing systems.