When the current anti-radiation technology and anti-radiation missile technology developed rapidly, active antiaircraft warning system can be easily found and attacked. Passive warning system is designed based on real-time scan-image processing by the optical imaging method to achieve region-wide early warning. A new parallel image processing structure is proposed to achieve real-time recognition of early warning air targets. At the same time, a new iterative identification algorithm is proposed based on Rough Sets (RS) and Support Vector Machine (SVM). The training data sets are separated into some blocks and their samples in sort-subjection are arranged to speed up the selection for support vector and the construction of the optimal interface. Therefore, training time of the sample set was greatly reduced and the recognition rate, generalization, realization performance of its recognition function was increased. Experimental results show that the system reached real-time early warning of air requirements while target recognition reaction time was about 15.1 ms and recognition rate was as high as 93%.