A Cross-grained Recognition Method For A Broad Variety Of Aircraft Sheet Metal Parts With High Similarity

It is necessary to paint the surface of aircraft sheet metal parts(hereafter referred to as SMPs)after forming.The classification and recognition of a broad variety of painted SMPs has always been done manually.There are many disadvantages such as inefficiency,inaccuracy and laborious when handling recognition task manually.This paper carries out research on cross-grained classification method based on machine vision among SMPs.The main research contents are as follow:1.Several image classification methods based on model training are compared and tested in the recognition task of aircraft SMPs.At first,SMPs' images data are collected from different perspectives in an unconstrained way.Image classification methods based on model training are implemented to recognize and classify the SMPs.Different image descriptors aimed at recognition and classification are researched.5 representative classifiers are constructed by combining descriptors like scale invariant feature transform(SIFT),histogram of oriented gradient(HOG)and classifiers like support vector machine(SVM)and multilayer perceptron(MLP).5 classifiers are implemented to recognize parts in SMPs sample set and the results are analyzed.2.A cross-grained recognition method based on machine vision is proposed.The recognition task of SMPs images is divided into 2 parts,which are recording of SMPs data and images recognition.The images recognition is also concluded by the coarse-grained and fine-grained step.Based on the analysis of different descriptors' performances,a 10-deminsonality coarse grained feature is constructed,which concludes shape descriptors and geometric invariants.The algorithm of nearest neighbor searching is used to complete coarse-grained classification.In order to continue the fine-grained classification task,the algorithms of matching based on contours and intensity information is tested separately.A guidance method based on visual information is also proposed in the fine-grained classification to ensure the stability of data acquisition and accuracy of classification.Aimed at the recognition method proposed in the paper,a set of hardware system is designed and constructed for capturing SMPs' images.After analyzing and choosing the types of equipment,appropriate industrial cameras,lens,illumination and other equipment are selected.The hardware system could take images from different points of view,which could reflect different shapes and structural characters of SMPs.Images with high quality could be acquired by rational distribution and adjustment.3.A set of software is developed to recognize and classify aircraft SMPs based on the broad algorithms research and comparison.The function,structures and logical framework is introduced in the paper.The abilities of the software are also tested.The cross-grained classification and recognition method proposed in this paper shows great performance in classification of test samples data.