Structure-context Based Fuzzy Neural Network Approach for Automatic Target Detection
基于结构上下文的模糊神经网络自动目标检测方法

This paper proposes a structure-context based fuzzy neural network(SCFNN) approach for automatic target detection. Fuzzy neural network methods not only possess advantages as adaptivity, parallelism, robustness, ruggedness, and optimality, but integrate advantages as depicting and solving system uncertainty by knowledge and rules of fuzzy set theory. Accordingly, they are powerful tools for image processing and pattern recognition. Use fuzziness measures as objective function of neural network can depict uncertainty of pixels' category validly so as to optimize image classification by minimizing the objective function. Puting information constraint of structure context on neurons' weighting process can reduce loss of image information, especially, the rich information comprised by target edges, by which target's attributes such as profile and shape can be retained validly, and the false detection rate can also be improved prominently. Experiments on remotly sensed images of target are executed to validate SCFNN approach. The results exhibit that SCFNN possesses good ability to automatic target detection, simultaneously, possesses valid abilities to eliminating uncertainty and retaining target shape compared with conventional neural network methods.