Satellite image classification is an important part of applications in various fields such as agriculture, environmental monitoring, and disaster management. K-means algorithm is a simple clustering method that can be adjusted for classification. Due to the fact that k-means represents a local search around the initially generated solutions, it should be combined with some global search method. We propose recent bare bone fireworks algorithm for k-means optimization used for image classification. The proposed method was tested on standard benchmark datasets and compared the results with other methods from the literature. Simulation results showed that the proposed combined approach is better for image classification compared to the original k-means algorithm, three other classification algorithms, and three methods based on other nature-inspired algorithms.