Study On Classification Spot Of Ablated By Femtosecond Laser Based On Multi-kernel Learning

When femtosecond laser technology is used to ablate a target object,the instantaneous high-density energy of the femtosecond laser can complete the ablation of the target object.When femtosecond laser technology is used to ablate single-crystal silicon wafers at microscale structures,plasma spots are often derived from the surface of silicon crystals.Due to the diffusion phenomenon in the outer halo part of the plasma spot image,the boundary contour of the spot image and its core area becomes blurred.Therefore,it is of great theoretical and practical significance to effectively and accurately filter out the outer halo portion of the plasma spot and obtain the core area of the spot image.The CCD cameras are used to collect plasma spot images under different ablation powers,and the relationship between different ablation powers and spot characteristics is analyzed.First,Image filtering should be used to eliminate the noise in the original spot image.Multi-scale Retinex enhancement method was used to enhance the spot image to highlight the transition layer between the spot core area and the halo.Then,the maximum inter-class variance method and human experience were used to segment the spot image to filter out some pixels that do not meet the expectations.During edge detection,the Canny operator could avoid the occurrence of breaks in the edges of the extracted spot image and obtain continuous contour features.Thirdly,the Freeman chain code was used to extract the spot contour,and the Hu invariant moment feature of the image and the normalized Fourier descriptor could express the contour feature of the spot image.Finally,the multi-core learning idea was used to construct a new composite kernel function.The normalized Fourier descriptor and Hu invariant moment were used as input samples to classify the spot image.Through comparative studies,it shows that when the kernel parameter ?=0.1,the penalty factor C is arbitrary,it is ideal to select the Hu invariant moment as the input feature;when the kernel parameter ?=0.5,?=1,the penalty factor C is taken as 100,it is ideal to select the Fourier descriptor as the input feature.Figure 36;Table 9;Reference 60...