A method for edge detection of full-depth convolution feature of standard parts

Abstract

A method for edge detection of full-depth convolution feature of standard parts, which includes training full convolution feature neural network, As an initial edge detection model (img file= 'DDA0001789553820000011. TIF' wi= '171' he= '95' / ), I2, I3... In... IN, where n is [1, N], n is Z, The images of standard parts were extracted from the original edge detection model (img file = 'DDA0001789553820000012. TIF' wi= '139' he= '88' / ), The standard part edge map (img file= 'DDA0001789553820000013. TIF' wi= '683' he= '91' / ) is obtained to compare the standard part image In with the standard partedge map (img file= 'DDA00017895 53820000014. TIF 'wi=' 125 'he=' 89 ' / ) Key points on standard part edge plot (img file=' DDA0001789553820000015. TIF 'wi=' 99 'he=' 89 ' / ) Edge (img file= 'DDA0001789553820000016. TIF' wi= '150' he= '96' / ) Non-critical edge (img file= 'DDA0001789553820000017. TIF' wi= '128' he= '93' / ) with the wrong edge (img file= 'DDA0001789553820000018. TIF' wi= '147' he= '94' / ) and set the critical edge (img file= 'DDA0001 789553820000019. TIF 'wi=' 115 'he=' 93 ' / ) as a positive sample, Non-critical edges (img file= 'DDA00017895538200000110. TIF' wi= '133' he= '88' / ) andincorrect edges (img file= 'DDA00017895538200000111. TIF 'wi=' 118 'he=' 88 ' / ) as a negative sample, Expert-aided edge map (img file = 'DDA00017895538200000112. TIF' wi= '700' he= '60' / ) is obtainedto reinforce the initial edge detection model. The new edge detection model (img file= 'DDA00017895538200000113. TIF' wi= '163' he= '97' / ) was obtained using the edge detection model (img file= 'DDA0001789553820 0000114. TIF 'wi=' 139 'he=' 97 ' / ) extract image edges of standard part, Obtain the standard part edge map (img file= 'DDA00017895538200000115. TIF' wi= '657' he= '88' / ) and judge whetherthe key edge is accurately extracted (img file= 'DDA000178955 38200000116. Re-execute Step C if TIF 'wi=' 700 'he=' 80 ' / ) and the extracted key edges can meet the requirements, if the requirements cannot be met; To meet the detection requirements, the (img file = 'DDA00017895538200000117. TIF' wi= '138' he= '98' / ) is used as the final edge detection model MEdge.

Description

technical field [0001] The invention relates to a deep learning edge detection method, in particular to a standard part full convolution feature edge detection method. Background technique [0002] Assembly refers to the process of connecting parts or components according to the specified technical requirements to make them semi-finished or finished products. Assembly is an important process of product manufacturing process, and the quality of assembly plays a decisive role in the quality of products. Extracting the edge of assembly standard parts is a key step in quantitatively evaluating assembly quality. Deep learning has achieved great success in image pattern recognition, classification and detection. If deep learning technology is applied to the edge detection of standard parts, edge detection will be improved. The cognitive ability of the method realizes the deep learning of artificial intelligence to the key edges on the standard part image, intelligently removes us...

AI Technical Summary

Problems solved by technology

It aims to solve the problem that the edge positioning in the image is not accurate enough, and the detected edge is not fine enough. Its core idea is to reconstruct the super-resolution image of the resulting image to achieve sub-pixel edge detection in the traditional sense; and the present invention is aimed at applying Artificial intelligence to distinguish key edges, the core idea is to use the importance of edges in the expert standard edge graph, the purpose, means, and final effect are all different

Images