Abstract: Solar radio burst data are crucial for understanding solar activity and forecasting space weather. Current deep learning-based detection methods primarily rely on dynamic spectra, which represent radio intensity, offering a limited dimensionality of information. To fully explore the physical information contained in spectral data, a detection method for solar radio bursts combining intensity spectra and polarization spectra is proposed. The approach begins with preprocessing raw data from the DAocheng Radio Telescope (DART) to directly generate high-quality image samples. And it utilizes a traditional object detection model to automatically identify the location of burst events and preliminarily distinguish the main bodies of four major types of radio bursts. Then, based on this, it is extended by constructing a polarization classification model utilizing ResNet18 that supports multi-channel input. This model can simultaneously process both intensity and polarization spectral information, ultimately outputting the event's type, time-frequency range, and circular polarization characteristics. The experiment results demonstrate that the model can effectively identify four common types of solar radio bursts. Furthermore, the multi-channel model (intensity channel and polarization channel) achieves comprehensively superior classification performance on polarization features compared to the three-channel pseudo-color image model that directly takes the polarization spectrum as input, confirming the value of polarization data as an additional feature dimension. This work provides an effective technical pathway for automatically extracting more physical information about burst events from massive volumes of solar radio data.