Research Of Application For Intelligent Image Recognition In Junior High School Geometry Automatic Marking

Machine automatic scoring is one of the important research fields of educational intelligence.However,only the automatic scoring technology of objective questions has been applied,and the automatic scoring of subjective questions in various subjects has not reached the practical level.The scoring of junior high school geometric subjective questions needs to comprehensively consider many problems such as logical reasoning,semantic understanding,drawing and auxiliary line addition results comparison.This paper mainly studies the extraction of global features of geometric figures and the method of geometric pattern recognition based on convolutional neural networks,and is applied to the graphics detection and identification business links in the junior high school geometric subjective review.The main research contents include the following points:(1)Research and implement the extraction algorithm of the edges and vertices of the geometric figures in the printed image.Taking the Hough line detection result as input,an edge extraction algorithm is designed to merge multiple line segments belonging to the same edge and output the endpoints of each edge in the geometry.Design the vertex extraction algorithm,extract the vertex coordinates and the associated edges of each vertex by acquiring the endpoints and intersections of each edge.(2)Research and implement the extraction algorithms for geometric topology.Based on the edge and vertex information in the geometry,the topology extraction algorithm is designed and implemented to obtain the adjacent vertices of each vertex on all its associated edges,and the geometric topology is described by the adjacency between the vertices.(3)Research and implement the isomorphic graph generation algorithm.According to the characteristics of isomorphic geometry,set the principle of isomorphic graph generation.According to the characteristics of the vertex associated edge and its own characteristics,the vertex mobility is judged,and the vertex moving range is determined.With the geometry topology as input,move each movable vertex to construct a homogeneous graphics library.(4)Study the construction of the isomorphic graphic training set.For the isomorphic geometry in the content(3),the image processing method based on contour detection is designed and implemented,and a square-shaped isomorphic figure with a fixed size is output under the premise of ensuring the geometric figure at the center of the image.On this basis,the isomorphic geometry training set is expanded by adding noise,rotation,affine and projection transformation.(5)Design and implement a hand-drawn geometric figure recognition algorithm based on convolutional neural network.Based on LeNet-5 model,a convolutional neural network with 4 layers of convolution and 4 layers of pooling is designed,and the isomorphic graphic training set in content(4)is used to train the model and realize the recognition of complex geometric figures in junior high school geometric subjective questions.Experiments show that the edge and vertex extraction algorithms designed in this paper have higher accuracy when detecting geometric shapes without corner marks.The topology extraction algorithm has higher accuracy when the edges and vertices are extracted correctly.The isomorphic graph generation algorithm based on vertex movement can introduce the image recognition algorithm based on convolutional neural network into the field of hand-drawn geometric pattern recognition by constructing a geometry graph library.The geometry recognition algorithm based on topology and convolutional neural network has high recognition rate and robustness and can be used to identify complex middle school geometry.