Research On Performance Evaluation For Target Recognition On Remote Image

With the rapid development of remote sensing technology, automatic target recognition (ATR) algorithms and systems are becoming more and more complex, how to improve their performance has been a prominent problem. Nowadays, the researchers of ATR technology are gradually wake up to the importance of performance evaluation for ATR, and the performance evaluation technology for ATR gradually becoming a hot research point. Contrapose to the problem of existing performance evaluation technology for ATR, researches will be done from the following aspects.(1) Performance evaluation indices for ATR are the basis of performance evaluation, the choice of indices affects the reliability of performance evaluation directly. Firstly, this thesis introduces several classical indices of ATR performance evaluation, and then according to the principle which is used for choosing evaluation indices and practical applications, the indices system of performance evaluation for ATR is designed.(2) There are so many subjective factors when deciding the weights of indices in existing methods of performance evaluation for ATR, contrapose to this problem, the BP neural network model is imported into performance evaluation for ATR. The BP neural network can find out the relationship between input and output via studying and training the sample data, so it can find the key of problem self-adaptively but does not depend on experience knowledge and rules to problem. This thesis emphasizes on lucubrating the methods of building model which is based on BP neural network to evaluate the performance of ATR, and discusses how to getting sample data.(3) Contrapose to the problem that performance evaluation for ATR is affected by image quality, a performance evaluation method for ATR which based on image quality is brought forward. In this thesis, we consult the method of national imagery interpretability rating scale (NIIRS) to grade the images in MSTAR dataset, make the MSTAR dataset into different subsets according to image quality levels. The targets in each subset are classed by the same ATR in order to examine the effects of image quality to the performance of ATR, and finally a performance evaluation method of ATR which based on image quality levels is brought forward. This method makes the performance evaluation results of ATR into the same standard, which can improves the robust of performance evaluation for ATR.