Objectives  Restoration of blurred aerial image can improve the details and features of images, and enhance the recognition capability and positioning accuracy of targets.

  Methods  A semi-blind restoration method of unmanned aerial vehicle image considering blurred kernel connectivity is proposed, which uses existing aerial images to establish the model of blurred kernel estimation under the conditions of non-parameters. Firstly, the gradient screening is established to filter out the public features of the blurred image and the existing clear image gradient domain, and construct the fidelity term. Then we use eight neighborhoods of blurred kernel gradient to describe the connectivity measurement of the blurred kernel and use it as a regular term to reduce the solution space and build the model. Finally, we reconstruct the image, estimate the blurred kernel hierarchically according to the image pyramid structure, and reconstruct the image by deconvolution with the split Bregman algorithm.

  Results  The experiment analyzes and evaluates the proposed method from four aspects: fuzzy type, public ground objects, comparison of methods and applicability of the method. Compared to the existing methods, the experiment results show that, when the public ground objects are over 35%, the blurred aerial images can be restored effectively.

  Conclusions  Compared with the existing image restoration methods, our proposed method has faster convergence speed, higher accuracy of fuzzy kernel, clearer details and richer information of restored image.