Implementation Of Vehicle Identification Algorithm On DSP Platform Based On Deep Learning

In the field of military reconnaissance,on the ground a variety of wheeled vehicles,crawler vehicles and other accurate reconnaissance and identification,easy to accurately assess its threat,real-time perception of the battlefield situation,master the initiative of the war.However,due to the particularity of the military vehicle database,this paper uses the civil vehicle database as the training set to design a real-time vehicle identification system with high universality,high recognition rate,better classification result and lower power consumption,and can be applied to any embedded Equipment.The design work of embedded model recognition system is divided into the following four aspects:(1)On the PC platform,a new type of vehicle is divided into pickup truck,van,SUV and MPV by using the depth learning algorithm to design a model with high recognition rate and better classification result.Sedan sedan,hatchback car six categories.(2)By successfully applying the vehicle identification model to the TMS320C6678 DSP chip,the embedded vehicle identification system is constructed.(3)According to the characteristics of DSP architecture,a series of optimization of embedded vehicle identification system on single core is carried out.Finally,the recognition time of single system is only 0.36 s.(4)Using the advantages of multi-core processor,the multi-thread parallel way is used to optimize the performance of the embedded vehicle identification system,and the recognition time of the final system is 0.15 s.Through the comparison and verification of the performance,the built-in embedded vehicle identification system has the advantages of low power consumption,real-time identification and so on.