Agricultural implement automatic navigation is a major trend of modern agriculture. Agricultural implement navigation can avoid the errors during implement shaking in field and improve the precision of navigation and flexibility of operation. At present, most of the navigation terminals were developed based on the industrial computer which is expensive and difficult to be spread. In this paper, an agricultural implement visual navigation terminal based on DSP and MCU for automatic weeding was designed. As the core processor of the system, DSP was responsible for the image acquisition, crop rows detection and offset calculation of navigation line. MCU is the main control unit of the system, so it was used to manage the working process, receive, store and forward GNSS data, and also control the implement. The corresponding protocol of serial communication, network communication and CAN bus communication in the system was normalized to make sure the stability of the communication. In the image processing, the OTSU method and the crop row detection algorithm based on boundary detection and scan-filter (BDSF) were adopted, which could improve the accuracy and efficiency of navigation line detection. In order to verify the validity and stability of the system, the algorithm adaptive test, offset accuracy test and different system comparison tests were carried out. The experiment results showed that the crop line detection algorithm could work adaptively in the weed and crop thinning conditions. The average error of offset detection is 1.29 cm and the maximum error is 4.1 cm. The systems comparison test verified the economic feasibility of the proposed system by compared with the PC and ARM, which can satisfy the requirements of filed operation.