The invention belongs to the field of pattern recognition and brain-like machine learning, particularly relates to an information processing method, system and device based on a contextual signal prefrontal-lobe-like network, and aims to solve the problems of complex system structure, poor flexibility and large training sample demand under complex conditions, namely complex multi-task conditions.The method comprises the following steps: selecting a corresponding feature vector extractor to carry out feature extraction; carrying out dimension matching on the information feature vector and eachscene signal in the scene signal set; and inputting the scene feature vectors after dimension matching into a feature vector classifier to obtain classification information. According to the method,a module similar to a prefrontal lobe is used for achieving multi-task learning facing context information, mapping depending on the context information can be learned step by step under the conditionthat the context information cannot be determined in advance, and the processed data can be applied to multi-task learning or continuous multi-task learning with higher requirements.