In recent years, the mental health problems of athletes have become more and more serious, and due to the complexity and invisibility of mental illnesses, it is difficult to detect and intervene in the early stage of mental illnesses due to a number of reasons. With the explosion of biomedical data and the rapid development of pattern recognition technology, the problem of recognizing mental illnesses can be solved by using artificial intelligence related technology. This paper proposes a mental illness recognition model based on Roberta Text CNN. Based on Roberta Text CNN model, a text classification model is built for two types of labels, namely, mental illness and suicidal tendency, and a comparison experiment is carried out with several mainstream models. The experimental results show that the proposed method achieves competitive results. In addition, this paper also introduces the EEG signals to be organized into dimensionally consistent matrices using Mel's inverted spectral coefficients for scale change, and then inputs them into the convolutional neural network as image features for further extraction of high-dimensional features and serial fusion of EEG features and then inputs them into the long and short-term memory network, and the application of the pattern recognition technology provides a two-branch mental illness recognition model, which further enhances the capability of screening and preventing early mental illnesses in athletes. The application of pattern recognition technology provides a two branch mental illness identification model, which further enhances the ability of screening and prevention of early mental illness in athletes