With the growth of the financial market, the securities market has become more important in the financial market, and it is conducive to the stability of the securities market to fully recognize the existence of risk and make preparations for risk prevention. The study proposes a generalized autoregressive conditional heteroskedasticity model based on particle swarm optimization algorithm for independent component analysis (PSO-ICA-GARCH), which is used for nonlinear volatility modeling of the securities market to generate a statistical image of securities market volatility. Then a multi-stage supervised bi-stream linear convolutional network (B-CNN)-based model is proposed for recognizing the images generated in the previous section to predict the rise and fall of the securities market. The results show that the PSO-ICA algorithm has higher separation accuracy compared to the ICA algorithm, and the four industries of agricultural services, environmental engineering, communication services, and logistics have significant positively correlated volatility spillovers to the electronics manufacturing industry, which verifies the existence of volatility spillovers among industries in the Chinese stock market. Using multi-supervised double convolutional neural network to analyze it to avoid the volatility of the stock market, the risk management of the stock market can be strengthened in this aspect.
