第5讲 免疫机器学习2.pptVIP

AIS Classification Results 5.3 免疫文件分类 Image classification using AIS The color images are those of the wooden components of kitchen cabinets Source : American Wood mark Corporation The images are obtained from Virginia polytechnic and state university Classification task involves two stages: a.Training of sample data : We have 8 sets of sample images The sample images are those of different colors and of different types of woods Each image is classified into a set of classes according to the type of wood and color b. Classification of test data 5.4 免疫图像分类 Types of WoodSource : American Woodmark Corporation ? ? Oak/frost Class - 2 Maple/frost Oak/Natural Class - 3 Oak/Honey Class - 4 Oak/Toffee Class - 5 Cherry/toffee Class - 6 Maple/white Class - 7 Maple/Natural Class - 8 Class - 1 5.4 免疫图像分类 The sets of sample images are used to train AIS The trained set of images are the antibodies The test images are the antigens that enter the body Training is experimented in two methods a. Matching algorithm –Negative selection of AIS b. Genetic algorithm Training 5.4 免疫图像分类 Matching Algorithm A random set of images are generated Each sample image is matched with the images generated and a matching value is found N images that have the minimum mismatch are chosen as the antibodies corresponding to that image The same process is repeated for all the sample images – This acts as the antibodies for classification Disadvantage : Limited number of antibody generation and hence limits the correct classification 5.4 免疫图像分类 Genetic Algorithm Generation of the initial antibody population - Corresponds to the sample image pool Find the fitness of the initial population with the sample images - Fitness function is the Euclidean distance between the RGB values of the sample image and the generated image Cross over the generated initial pool to get the children pool of images Repeat the process and generate the children pool till