Calculation Of Area Based On Image Segmentation And Reserch On Semantic Segmentation Algorithm

With the continuous reform and innovation of information technology,the Internet has become a main way for people to obtain information,and data information can be presented in a variety of expression forms.Image is one of the most important carriers for carrying information,how to obtain necessary information from massive image data,and its processing method,that is,has received extensive attention and research from the majority of researchers in recent years.Many common methods of image processing include image transformation,image compression,image enhancement,image restoration and image segmentation,etc,among which the image segmentation plays an important role in image processing.There are many methods of image segmentation,and it is particularly important to select the appropriate segmentation method for specific objects.Therefore,for the image segmentation problems,the research contents are as follows:1.In the past,when the point electrode area in the MASK was calculated,the point electrode region was directly regarded as a circle,and this calculation method had large numerical error.Therefore,the region growing method is used to segment the effective point electrode region,and then calculate it.The experiment results show that the segmentation accuracy of this method is relatively accurate,but there are some problems of incomplete segmentation or excessive segmentation in some regions.2.Aiming at the problems in the aforementioned research methods,a methods of image segmentation based on the combination of clustering segmentation and region growing is further proposed.Firstly,the image is segmented by the clustering,and then the feature image is segmented by the region growing,finally,the divided electrode are calculated the numerical area.The experimental results on the same data sets show that the segmentation accuracy has been significantly improved,and the problems of incomplete segmentation and excessive segmentation are effectively solved.3.The segmentation result images in method 2 are used as the labels,the original images and the segmented result images are cut with individual point electrode images as data sets,and such images were applied to FCN and UNet network and the electrode area is calculated.The experiment result shows that the electrode area is regular and the edges are smoother.The accuracy of the segmentation is high.4.An improved image semantic segmentation algorithm is also proposed based on FCN.This method aims at the problems of inaccurate segmentation accuracy and loss of image details existing in traditional FCN,this method uses dilated convolution instead of standard convolution to expand the receptive field area;uses variable pooling to capture dense information as much as possible;uses cascade operation to fuse deep and shallow information.The public data set is that the remote sensing satellite observe roads.The experimental results show that,compared with the traditional FCN,the accuracy of segmentation has been effectively improved.