Abstract: ‍ ‍Waves cause constant changes in the signal power of sea clutter， which demonstrates obvious nonhomogeneous characteristics. The detection performance for maritime weak targets is seriously restricted owing to the reduced estimation accuracy of clutter. To solve this problem， many detection algorithms have been designed to improve the sea-clutter detection performance. The information geometry detector， which maps the received signals to a matrix manifold and makes decisions based on the geometric distance in the manifold space， has shown great advantages in weak-target detection. However， its performance still suffers from the poor estimation accuracy of the clutter covariance matrix. To further improve the performance of the information geometry detector in sea-clutter environments， a maritime weak target detection method based on a manifold filter is proposed herein. First， the signals of the range cells were modeled as Toeplitz positive definite matrices， forming a matrix manifold. Each range cell corresponded to a point on the manifold， and the characteristics of the sea clutter could be expressed by the geometric topology characteristics of the manifold. Specifically， the points of range cells with similar statistical characteristics were closer on the manifold， while the nonhomogeneous points were separated. Next， considering the geometric structure of the manifold space， a clustering method was proposed for the manifold. A nonhomogeneous clutter was divided into several types of clutter with similar characteristics using this clustering method， and the geometric centers of these kinds were estimated. Because the clutter in a class demonstrated closer geometric distances， it was appropriate to use their geometric centers to represent the geometric distributions of these classes. Finally， based on the number of samples contained in each class， the weights of each clutter class were calculated to reduce the influence of nonhomogeneous clutter. Then， the manifold filter was used for different types of clutter to improve the estimate accuracy of the clutter covariance matrix. Experiments based on both simulated and measured data were conducted to analyze the performance of the proposed method and compared methods under different scenarios. The results showed that the proposed method could effectively improve the detection performance for maritime weak targets in sea-clutter environments.