1. Surface Defect Inspection: an Artificial Immune Approach Dr. Hong Zheng and Dr. Saeid Nahavandi Email: hzheng@deakin.edu.au School of Engineering and Technology Deakin University, Geelong, VIC 3217, Australia 2002

2. Outline 4 Introduction to texture surface inspection 4 Aim of this research 4 Brief outline on Artificial Immune System 4 Clonal selection principle 4 Clonal selection algorithm (CSA) 4 Learning defect inspection using CSA 4 Experimental Results 4 Conclusions and future Work

3. Introduction to texture surface inspection 4 Methodology –structural approach The method analyzes a texture by describing both the primitives and the placement rule. It is suitable for regular texture analysis. –statistical approach The method describes a texture by a probability distribution model. It is suitable for regular and irregular texture analysis. 4 Drawback –Unable to adapt to the changes in orientations, scales and shapes of defects. –

4. Aim of this research Study a novel methodology of texture defect detection, which is able to extract robust texture features invariant to the changes in orientations, scales and shapes of defects.

5. Immune System 4 Nature immune system –Main functions Protect our body from infectious agents such as viruses, bacteria, fungi, and other parasites. –Basic components lymphocytes or the white blood cells 4 Artificial immune system A novel computational intelligence paradigm inspired by nature immune system.

6. Immune system architecture Figure 1. Immune system architecture Phagocyte Adaptive immune response Lymphocytes Innate immune response Biochemical barriers Skin Pathogens

7. Clonal selection The clonal selection is the theory used to explain how an immune response is mounted when a non-self antigenic pattern is recognized by a B-cell. It consist of three main processes:selection, proliferation and affinity maturation.

8. The clonal selection principle M M Proliferation and Mutation Selection Foreign antigen High affinity memory cells Antibody Figure2. The clonal selection principle

9. Clonal selection algorithm (CSA) The clonal selection algorithm is proposed to fulfil above clonal selection processes. This algorithm was initially proposed to perform pattern recognition and solve multi-modal optimization tasks. Define a set of patterns to be recognized and call it the non-self set (P). Based upon the clonal selection algorithm, generate a set of detectors (M) that will be responsible to identify all elements that belong to the non-self set. The algorithm as summarized by Castro is as follows :

10. Clonal selection algorithm 1. Randomly initialize a population of individuals (M); 2. For each pattern of P, present it to the population M and determine its affinity (match) with each element of the population M; 3. Select n1 of the best highest affinity elements of M and generate copies of these individuals proportionally to their affinity with the antigen. The higher the affinity, the higher the number of copies, and vice-versa; 4. Mutate all these copies with a rate proportional to their affinity with the input pattern: the higher the affinity, the smaller the mutation rate, and vice-versa. 5. Add these mutated individuals to the population M and re-select n2 of these maturated (optimized) individuals to be kept as memories of the system; 6. Repeat Steps 2 to 5 until a certain criterion is met, such as a minimum pattern recognition or classification error.

11. Learning defect inspection using CSA 4 Feature extraction 4 Learning inspection parameters

12. Feature extraction 4 The principle texture statistic utilized to represent the texture feature is the normalized “texture energy” derived from Law’s approach: the standard deviation of pixel gray scale within a N*N window size computed after convolution with an optimal texture filter through task-aimed training based on clonal selection algorithms.

13. Learning inspection parameters using CSA 4 Definition of immunological terms –Antigen Any of training textile texture images. –Antibody A float string encoded by filter parameters and a segmentation threshold. –Affinity The percentage of correct detection of a antibody. It is defined as following:

14. Learning inspection parameters using CSA 4 Immune evolutionary operation –Clone : This operation is to generate copies of every individual in an antibody population proportionally to its affinity with the antigen. All individuals are sorted in descending order firstly. The amount of clones of a antibody is given by Where N is the number of all individuals in an antibody population. f i is the affinity value of the ith antibody.

15. Learning inspection parameters using CSA 4 Immune evolutionary operation –Mutation: The mutation operation creates a new antibody by randomly changing one or more of the unit values in the antibody with a probability proportional to their affinity. The mutation probability is given by Where f max is the highest affinity value, and f min is the lowest affinity value.

16. Learning inspection parameters using CSA 4 Immune evolutionary operation –Reselection: This operation sorted all individuals in descending order, and replace n1 of the lowest affinity antibodies with n1 new randomly generated antibodies.

17. Learning algorithms description Create an initial antibody population Calculate antibodies’ affinities Clone operation Generations>n ? Mutation operation Reselection operation Optimal antibody N Y

18. Defect detection procedure based on the optimized antibody Input detected textures Detection Image filtering Image threshold Defect textures Defect-free textures Optimal antibody (A filter and a threshold)

19. Experimental Results Texture learning samples from TILDA database (a) Defect texture image sample (b) Defect-free texture image sample Figure 3. Textile image sample set A (a) Defect texture sample images (b) Defect-free texture sample images Figure 4. Textile image sample set B

20. Experimental Results Defect inspection results

21. Conclusions and further work 4 Conclusions Nature immune mechanism can be used to build novel computational tools to solve difficult problems in the field of visual inspection. 4 Further work Further explore the same approach and find general principle to detect various surface defects in industry.

22. Thank You