Humoral Cell Recognition Based On BP Neural Network

The research for automatic recognition and analysis of cell image is quite active in the domain of biomedical engineering. Automatic recognition and analysis of humoral cell image's corporeal components can improve the efficiency of assay and the results can be more scientific. With the development of Artificial Neural Network(ANN) and image science, image recognition based on ANN is much better than traditional pattern recognition. The application of image recognition can be widely used in many fields. This thesis mainly includes three aspects: image acquisition and preprocessing, extracting image combination moment, establishing the BP neural network model to recognize different target images.(1) The image database has been built by image acquisition system and the images have been processed by image denoising, segmentation and feature extraction. By analyzing degraded humoral factors and cell image noise model, the adaptive median filtering algorithm combining information entropy noise detection has been proposed, and then the best threshold gray image segmentation algorithm and the feature extraction algorithm based on combination moments have been given.(2) 5 Combined-moment invariants which based on 7 Hu moment invariants have been presents in the paper, and then the combined-moment invariants are extracted from binarization images by proposed program.(3) The basic principle of BP neural network algorithm used for cell recognition is introduced. After been comprised with the traditional pattern recognition, neural network classifier is selected for image classification and recognition systems, and used for to classify and identify the target image. The algorithm includes two steps: training and testing. In the training step, the BP neural network model is determined, and the parameters of BP neural network are trained by extracting the image features. In the testing step, the neural network model and parameters which have been trained are used to identify and classify the cell image.Finally, 200 experiments using BP neural network algorithm for recognition and analysis of cell image is given. These results show that the algorithm is more effectively for automatic recognition and analysis of cell image.