The current mainstream operation methods are mainly oriented to cooperation goals and pre-designed signs, which are difficult to meet the needs of complex tasks. Taking the non-cooperative target as the research object, this paper proposes a method of active light source detection combined with different multi-source data processing networks. When the image brightness is relatively high, the residual convolutional neural network is used to identify the target capture area in the multi-source data; when the image brightness is dark, the PointNet++ network is used to identify the capture area from the point cloud data, thereby adaptively achieving a better segmentation effect. The final experiment proved the effectiveness of this method.