Fluid inclusions are carriers that record information on fluid activity within the Earth. By analyzing the parameters of fluid inclusions, we can gain a deeper understanding of fluid movement within the Earth and reveal the distribution patterns of mineral resources. The parameter analysis of traditional fluid inclusions is based on human observation and description, which has strong subjectivity, can only be qualitatively evaluated, and cannot be quantitatively analyzed. This article uses deep learning and image processing methods to quantitatively analyze the parameters of fluid inclusions. In order to accurately obtain the content of each component and clear contour information, an improvement was made on the Unet semantic segmentation model. An attention gating module was added to the skip connection part of the network. In the decoding part of the network, two layers of feature maps were selected for each scale. After upsampling, feature maps of different scales were uniformly sized and fused to achieve accurate segmentation of fluid inclusions. By combining digital image processing technology and OCR, the aspect ratio and gas-liquid ratio of fluid inclusions can be calculated, achieving intelligent analysis of fluid inclusion parameters and providing effective data support for fluid inclusion research.