An offline oil detection device is designed based on the heterogeneous architecture of a System on a Programmable Chip (SoPC). The SoPC platform is constructed by porting Cortex-M0 soft core to Field Programmable Gate Array (FPGA). Using FPGA hardware resources, the operation order is optimized through the pipeline design with serial-parallel hybrid methods. Moreover, the parameter calibration accelerator and neural network accelerator are designed and connected to the bus system. Experimental results show that this device can perform the qualitative analysis and classification of gutter oil accurately with high detection efficiency. Compared with the traditional similar terminal gutter oil detection equipment, the detection time is shortened by 89%. Moreover, the data processing speed of SoPC increases by twice comparing to Cortex-A9, with the introduction of accelerator.