Abstract: Compared with the traditional machine learning algorithms, deploying deep neural networks in embedded systems can significantly improve the performance of robot system object recognition. However, its performance is limited by the computing resources and memory capacity of the embedded platform. Hence, it is quite necessary to simplify the network structure and improve the system efficiency through model pruning and parameter quantification. Meanwhile, it is also necessary to prevent overfitting through dropout regularization and improve the accuracy of system recognition. In order to further improve the object recognition performance of the deep neural network algorithm in the embedded robot system, this paper proposes a deep neural network dropout regularization method based on constant false alarm detection (CFAR-Dropout). Firstly, by quantizing the weights, the weights and activations are reduced from floating point numbers to binary values. Secondly, a constant false alarm detector (CFAR) is designed to maintain a certain false alarm rate, adaptively delete some neuron nodes, and optimize the neuron nodes involved in the calculation. Finally, on the embedded platform PYNQ-Z2, an VGG16-based optimization model is used to experimentally verify the object recognition performance of the proposed algorithm. It is shown via experimental results that compared with the classic dropout regularization method, the CFAR-Dropout regularization method can reduce the error rate by about 2%, and effectively prevent the overfitting. Moreover, compared with the original network structure, the proposed algorithm can reduce the amount of the occupied memory by about 8%, and effectively prevent over-parameterization.