Considering that it's not easy to detect the defect of the pixels on the edge of hyperspectral image with slide window, we used edge expansion method to make the pixels on the edge be detected normally. Then, a new hyperspectral data model was established in the form of local summation in slide window, hence the rationality was promoted. At last, to prove the effectiveness of the proposed algorithm, experiments were implemented with real data. The results show that this algorithm performs better than the traditional algorithms of RX and KRX, and it can detect more anomalies with lower false alarm rate.