Target Recognition Technique Research Based On Fuzzy Logic Control

Fuzzy logic control system, as an intelligent learning system, comprises of fuzzifier,rules, fuzzy inference engine and output processing four main modules. The employment of the soft-decision and learning ability of fuzzy logic system in target recognition and parameters’ recognition tasks will fully utilize the related information embedded in the received raw data, which is able to enhance the ability of recognition. When embracing the big data era, the combination of fuzzy logic control and recognition will boost more research directions and techniques. However, there are little literatures that cover the contents relevent to this topic. In this work, two types of fuzzy logic systems are adopted in two typical target recognition applications. It includes the followings:1. The theory of two main fuzzy sets and the corresponding properties and operations are introduced. It contains the definition of membership function, operations of type-1 fuzzy sets and type-2 fuzzy sets, and the computation between the same set and different sets, etc. The framework of type-1 fuzzy logic system(T1FLS) and type-2 fuzzy logic system(T2FLS), the design of related modules in the framework and logic expressions are illustrated, which discloses their relation. Furthermore, the interval type-2 fuzzy logic system is also introduced.2. The T1 FLS model is applied to recognize the high range resolution profile(HRRP) of the air targets generated from the echoes from the wide band radar. The model is aimed to fuse the two characteristics obtained from the HRRPs of different types of air targets. Firstly, the categories of targets are classified based on the “experts’ experiences”.Then, the membership functions corresponding to the fuzzifier are designed based on the experiences. Similarly, we design the rules and defuzzifier. The data of two features obtained from the HRRP are imported in the pre-defined two types of fuzzy sets. After the process of fuzzifing, inferring via rules and defuzzifing, the corresponding type of target is recognized. One step further, the recognition performance under different probability of false alarm and signal to noise ratio conditions are compared.3. The T2 FLS is studied in order to handle the high uncertainty caused by the randomness of the UWB signal and noise, and it is compared with the T1 FLS. Two systems are employed in the UWB radar measurements–echoes of two types of soil with different water contents. Water is poured into the two types of soil to adjust different VWC values. Calibrated with the auxiliary time domain reflector, large amount of data are collected for each cases. After preprocessing, multiple echoes within the same case are concatenated to form an aperiodic signal for time series modeling. Then, T1 FLS and T2 FLS are employed to forecast and learn. Iterated by the back-propagation method and modified K-M method respectively, when the root mean square error of the prediction and real data converges under a threshold, the parameters of the membership functions are extracted as the templates of different cases of two soil types. Using a template matching strategy to process the unrecognized echoes, the VWCs are obtained. Furthermore, we compare the recognition performance under two different criteria.The results of the two application problems are two-fold basically:(1) For air targets recognition, compared with two different types of design in membership functions in fuzzifier, it is better for F-15 and AH-64 using the combination of triangle and trapezoid membership functions than using the gaussian membership functions, but for Tu-16 the result is opposite. Therefore, the design of fuzzy logic system is affected by the target characteristics and the knowledge of experience.(2) For soil’s VWC content recognition, the time-series forecasting of the two fuzzy logic systems are converging and the templates of parameters are obtained, which correponds to the membership functions’ coefficients. Recognition results demonstrate that the performance of IT2 FLS is better than the T1 FLS due to the consideration of background noise, which contains a higher level of uncertainty, but the computation cost is higher for IT2 FLS.