The invention discloses a convolutional neural network based video retrieval method and system. The method comprises the steps of firstly constructing a convolutional neural network based computing model; training the computing model through image data to obtain an optimized computing model; removing a classifier in the model to obtain an extraction computing model; performing transition frame extraction on an existing video resource, extracting transition characteristics of transition frames through the extraction computing model, and establishing a transition characteristic database; and performing transition frame extraction on a to-be-retrieved video to obtain transition characteristics, and retrieving the transition characteristics in the transition characteristic database to obtain a retrieval result of the video. According to the convolutional neural network based video retrieval method and system, the transition frames in the video serve as data processing objects, so that the robustness of the retrieval process is improved and redundant information is removed; and the characteristics are extracted through the convolutional neural network based computing model, so that the video retrieval speed is increased and the video retrieval accuracy is improved.