In finite element (FE) modelling, the visual assessment of images of contour plots is widely used for qualitative comparison and analysis. However, this approach has limitations in conducting in-depth and comprehensive analysis, which is rigorously made based on objective, massive, and quantitative data. To tackle the problems, we propose a methodology based on computer vision with k-means clustering algorithms, providing an effective quantitative strategy to improve the efficiency and accuracy of the analysis. To demonstrate the application of this methodology, three levels of FE models on impact, viz. macro (homogenous UD laminate model), meso (yarn-level fabric model), and micro (fibre-level yarn model), are developed, where the images with both simple and complex morphologies and backgrounds are all be considered. The contour plots of stress and transverse deformation are clustered into k categories based on the RGB values of the pixels in the images. Notably, the quantification of one image using the proposed methodology is completed in seconds, and the efficiency is enhanced by approximately 40 times compared with that using the manual method, without the accuracy sacrificed. This methodology enables the researchers to efficiently and precisely understand the ballistic behaviours of the textile materials from the perspectives of fibres, yarn, fabrics, etc., and has great potential to apply in other research fields, where computational simulation is widely used.