To address the problem of insufficient recognition accuracy for uplink signals in two-way automatic communication systems（TWACS）, a signal recognition algorithm based on the dandelion optimization（DO） algorithm that combines convolutional neural network（CNN） with attention mechanism（AM） and bidirectional long short-term memory（Bi-LSTM） neural networks is proposed, which is briefly referred to as the DO-CAB algorithm. The algorithm first adaptively extracts important features of TWACS signals using a CNN. It then optimizes the hyperparameters of the Bi-LSTM using the DO algorithm, constructs the network based on the optimized hyperparameters, and introduces an AM to assign influence weights to the inputs, improving the network algorithm for better signal recognition. The experimental results show that the proposed algorithm achieves a recognition accuracy of 92.32%, enabling efficient and accurate identification of TWACS modulated signals.