The embedded system based on FPGA+DSP was suitable for the agricultural sorting field, due to its flexible chip selection and high power efficiency. The embedded system was designed based on the principle of detection, software and hardware design and rejection system design. In principle of detection, a polarization channel was added to the UV channel to resolve the problem of transparent film detection without increasing camera. In software design, a speed measurement based on spatial correction and a learning algorithm for threshold detection was proposed to improve the adaptation of the equipment. An embedded solution was put forward which had advantages in both material cost and power consumption. After optimization, the DSP embedded board could stably process camera image data in real time. In rejection system, a control scheme was designed based on target real-time speed. The experiment result showed that the speed measurement scheme was verified via roller platform, and cotton speed can be obtained in the production line test. The embedded system could meet the real-time requirement, and the system was stable under varying foreign fiber numbers in long-term test. Concerning the test of rejection system, an impact test was performed by changing the wind speed. Finally, two typical foreign fiber detection tests of the system were implemented. The results showed that the detection rates of foreign fibers and polypropylene filaments as well as transparent films were higher than 80%, while the detection rate of yellowish foreign fibers was slightly lower than 80%. In the comparison test with similar equipment, the present equipment revealed superior detection rate. Long-term test result showed that the present equipment was easy to operate and had stable performance.