At present, countries around the world are vigorously carrying out research on key technologies for manned/unmanned aircraft coordination, but there is still a large risk in carrying out manned/unmanned system test flight tests due to the current level of intelligence of unmanned equipment. Therefore, there is an urgent need to carry out research work around the online assessment method of collision risk in cooperative test flight. In this paper, we first use the trajectory prediction model based on CNN algorithm to train the position and attitude of the aircraft as the input of the trajectory prediction model, which is more in line with the change rule of the real trajectory; after that, we build the trajectory dispersion cone based on the predicted trajectory data and the deviation range; finally we combine the spatial geometric relationship to construct the collision risk detection model, which realizes whether there is a collision risk in the future period of time. Prediction. The results show that the trajectory prediction model based on CNN algorithm has better accuracy; and the prediction results of the flight conflict risk detection model basically coincide with the previously set flight trajectory conflict situation, which verifies that the model in this paper is effective and able to accurately capture moments of potential collision risk, and carry out on-line safety assessment of manned/unmanned aircraft coordinated test flights.