The Research Of Handwritten Numeral Recognition Based On Generalized Regression Neural Network

In recent years, the handwritten numeral recognition not only has been a research hotspot for researchers in national or aboard, but also become one of the most successful applications in the field of pattern recognition and it has a wide range of applications in real life. However, handwritten digits have a large arbitrariness due to the writer, so there is plenty of room for improvement in the recognition accuracy of handwritten.The major work of this paper is to study the handwritten numbers with neural network. Firstly, use a series of numeral images pretreatment to realize image grizzled processing, smooth denoising, binarization, elaboration and normalization. Secondly, describes several common feature extractions methods and adopts the coarse mesh feature extraction method which extracts the eigenvectors from the processed handwritten numeral image with the coarse mesh feature extraction for next research. Finally, compare the radial basis function neural network (RBFNN) with generalized regression neural network (GRNN) to study handwritten numeral recognition, then use drosophila optimization algorithm (FOA) to optimize the generalized regression neural network expansion constants (SPREAD) and carry out a further study to compare the drosophila algorithms which has optimized-general regression neural network (FOA_GRNN) with generalized regression neural network (GRNN).As the experimental results suggest, not only does the computational speed of the generalized regression neural network(GRNN) method is faster than the radial basis function neural network (RBF) method but also the digital recognition is more accurate and a network of four general regression neural network could output larger results than the radial basis function neural network in hidden layer node when the input signal is near the center of the base function. It can be concluded that the former has advantages on the approximation ability and learning speed over the latter and its handwritten numeral recognition is faster and more accurate. Compared with the first two methods, the general regression neural network after optimization still has much scope to improve the recognition rate and convergence rate.