Abstract: During the testing of aircraft airborne systems, the number of collected test images is large and the types are random, and it is impossible to find relevant images based on the characteristics of the target image itself. In this paper, a lightweight unsupervised image clustering retrieval method is designed based on the improved K- means clustering algorithm. The method split large sample images, and the principal component analysis (PCA) was introduced for data dimensionality reduction, then the obtained sample vectors could be quickly clustered. Furthermore, a Gaussian weighted voting and clustering probability similarity mechanism were designed to calculate the clustering probability similarity between samples. Based on the breadth first search (BFS) algorithm, the similarity samples were extended to complete image retrieval. The experimental results show that the method can effectively classify the test image categories and can quickly retrieve relevant images according to the image features, which meets the online retrieval requirements for aircraft airborne system test images.