Abstract: The propagation of methane gas in coal is closely linked to the particle size distribution characteristics of coal particles, which in turn affects the safe mining and utilization of coal. With the continuous development of digital image processing technology, coal particle morphology detection based on digital image segmentation has become the mainstream method for obtaining the particle size distribution characteristics of coal particles. In the process of digital image segmentation, global information and edge details play a crucial role and directly affect the accuracy of the segmentation results. The U-shaped network based on convolutional neural network architecture focus too much on local information, neglecting the importance of global information, which can lead to over-segmentation. On the other hand, Transformer-based networks effectively model global information using multi-head self-attention mechanisms but do not fully utilize edge detail features, resulting in under-segmentation of coal particles. To address these issues, this study proposes an Iterative Squeeze UNet (ISUNet) for coal particle size analysis. The ISUNet model introduces a compressed excitation atrous spatial pyramid pooling module and a Transformer-based multi-path iterative encoder. The compressed excitation atrous spatial pyramid pooling module enhances channel information and global context information of features at different scales, solving the problem of over-segmentation of coal particles. The multi-head self-attention module in the encoder continuously strengthens important edge detail features through dot-product self-attention mechanism, addressing the problem of under-segmentation of coal particles. Compared to five classic image segmentation models and four mainstream segmentation models, ISUNet has shown outstanding performance. Compared to the classic segmentation model TransUNet, it has improved the mean Intersection over Union (mIoU) by 6.6%, the accuracy by 0.3%, and the recall rate by 7.0%. Compared to the current state-of-the-art Segment Anything model, it has improved the mIoU by 4.6%, the accuracy by 0.2%, and the recall rate by 4.9%. In the aspect of coal particle size measurement, the accuracy reached 97.49%. These experimental results fully demonstrate the effectiveness and superiority of ISUNet in coal particle size analysis.