In ground-based astronomy observation

overcome image blurring caused by atmosphere turbulence is key to obtain high-resolution astronomical images. Deconvolution is an important step in process of image restoration

and deconvolution algorithm based on scaled gradient projection is suitable for astronomical image restoration with high recovery precision

processing efficiency

and good robustness. But in practice

the clear images are unknown and interferences caused by noise exist

the algorithm is unable to stop iteration

making its application limited. Combined with no-reference quality measurement

an improved algorithm which can automatically stop iteration is proposed. Simulation indicates that the improved algorithm can accurately judge the optimal point and can be applied to real astronomical image processing.