In view of the fact that the traditional threshold segmentation algorithm relies on the distribution model of the background clutter severely and its poor performance in noise resistance and robustness, such an algorithm as λ-detection is proposed, which is based on the improvement of the local signal-to-noise ratio model in the traditional energy detection, and the problem that the threshold cannot be adaptively selected is solved. Aiming at the problem that the algorithm has insufficient processing power when there are a large number of speckles and smear in the synthetic aperture radar (SAR) image, the neighborhood pixel mean is considered to extend it to two-dimensional, then the threshold segmentation method based on the two-dimensional energy detection comes out. Finally, by introducing three indicators of uniformity of intra region, dissimilarity of inter region and shape complexity, compared with the popular maximum entropy threshold segmentation and improved 2D maximum inter-class difference method, the proposed algorithm has been proved simple and effective.