In order to improve the detection and recognition ability of wireless communication systems in complex channel environments, and the problems of complex models, large calculation amounts, and incomplete input data characteristics. We propose an improved deep learning algorithm model to identify nine common modulated signals in real wireless environments. The algorithm calculates the amplitude and phase of the original in-phase quadrature (IQ) data to increase the characteristic information of the input data of the model and uses an improved dense neural network (DenseNet) to identify and classify common modulation signals. The experimental results show that the proposed algorithm has the best performance compared with other deep learning modulation recognition algorithms in the same training data samples. At 0dB, the average recognition rate of DenseNet reaches 84.6%. The improved IQ input data significantly improves the detection and recognition rate of the wireless signal. When the SNR is -10dB and -5dB, the recognition rate of the modulated signal is increased by 10%.