Using Back Propagation Algorithm and Genetic Algorithm to Train and Refine Neural Networks for Object Detection

We introduce a two stage approach to the use of pixel based neural networks for finding relatively small objects in large pictures. In the first stage the network is trained by using a back propagation algorithm and a genetic algorithm on sample objects which have been cut out from the large pictures. In the second stage the weights of the two trained networks are refined on the full training images using a second genetic algorithm. Four methods are formed by the two training algorithms in stage one and combining them with the genetic algorithm in stage two. We have tested these methods on three object detection problems of increasing diffculty. In all cases the methods with the refined genetic algorithm resulted in improved detection performance over those without the refinement. The method with the best performance is the use of the back propagation algorithm for the network training and the genetic algorithm for the network refinement.

[1]  PAUL D. GADER,et al.  Segmentation free shared weight networks for automatic vehicle detection , 1995, Neural Networks.

[2]  W. Spears,et al.  On the Virtues of Parameterized Uniform Crossover , 1995 .

[3]  M. W. Roth.  Survey of neural network technology for automatic target recognition , 1990, IEEE Trans. Neural Networks.

[4]  Geoffrey E. Hinton.  Connectionist Learning Procedures , 1989, Artif. Intell..

[5]  Victor Ciesielski,et al.  An evolutionary approach to training feedforward and recurrent neural networks , 1998, 1998 Second International Conference. Knowledge-Based Intelligent Electronic Systems. Proceedings KES'98 (Cat. No.98EX111).

[6]  Walter Alden Tackett,et al.  Genetic Programming for Feature Discovery and Image Discrimination , 1993, ICGA.

[7]  D. E. Goldberg,et al.  Genetic Algorithms in Search , 1989 .

[8]  Abdesselam Bouzerdoum,et al.  The Evolutionary PreProcessor : Automatic FeatureExtraction for Supervised Classi cation using GeneticProgrammingJamie R , 1997 .

[9]  Paul D. Gader,et al.  Morphological shared-weight networks with applications to automatic target recognition , 1997, IEEE Trans. Neural Networks.

[10]  Alan N. Gove,et al.  Neural processing of targets in visible, multispectral IR and SAR imagery , 1995, Neural Networks.

[11]  Manuela M. Veloso,et al.  A Contolled Experiment: Evolution for Learning Difficult Image Classification , 1995, EPIA.

[12]  David P. Casasent,et al.  Classifier and shift-invariant automatic target recognition neural networks , 1995, Neural Networks.

[13]  James L. McClelland,et al.  Parallel distributed processing: explorations in the microstructure of cognition, vol. 1: foundations , 1986 .

[14]  Geoffrey E. Hinton,et al.  Learning internal representations by error propagation , 1986 .

[15]  Jin Wang,et al.  Weight smoothing to improve network generalization , 1994, IEEE Trans. Neural Networks.

[16]  Mohan M. Trivedi,et al.  A neural network filter to detect small targets in high clutter backgrounds , 1995, IEEE Trans. Neural Networks.

[17]  Mengjie Zhang.  A Pixel-Based Approach to Recognising Small Objects in Large Pictures Using Neural Networks , 1997 .

[18]  Dana H. Ballard,et al.  Computer Vision , 1982 .

[19]  Herbert L. Roitblat,et al.  Sonar recognition of targets embedded in sediment , 1995, Neural Networks.

[20]  Bangalore S. Manjunath,et al.  Genetic Programming for Object Detection , 1996 .

[21]  Lashon B. Booker,et al.  Proceedings of the fourth international conference on Genetic algorithms , 1991 .

[22]  David E. Goldberg,et al.  Genetic Algorithms in Search Optimization and Machine Learning , 1988 .