Research On CAD Graphic Recognition Based On Random Forest Algorithm

In the early stage,the budget for the price of construction projects completely depended on the cost engineer to manually enter the unit price of construction materials into the calculator one by one by using the calculator for summary calculation.This process is not only cumbersome,time-consuming and labor-consuming,but also is very easy to get wrong.After the 1990s,computer technology and network technology have made great progress.Some engineering cost software,engineering quality management,engineering quantity calculation and other software have also emerged in the construction field.With the wide application of engineering cost and quantity calculation software,cost engineers get rid of calculators.With the attack of the wave of artificial intelligence,the application and research of artificial intelligence in the construction field are in full swing and booming.The industry has carried out extensive discussion on how to design CAD graphic recognition business and other related topics.The key step of the project cost is to identify the CAD elements in the drawings through engineers,then model and calculate the elements,and finally complete the cost calculation.Therefore,how to recognize CAD primitives more accurately has always been a research focus of experts,scholars and practitioners in the field of drawing recognition.In order to effectively complete CAD graphic recognition and improve the recognition accuracy,the main work of this paper is as follows:Firstly,the characteristics of architectural engineering drawings and CAD primitives are analyzed,the characteristics of common CAD primitives are introduced,and the basic principle of YOLO algorithm for target detection is expounded.The corner function is introduced to solve the problem of data balance,the target detection algorithm is used to locate the target image of CAD primitives,and the Non-Maximum Suppression algorithm is used to optimize the target detection result and extract the target image.Secondly,the hog algorithm and Harris algorithm,which have good description ability for CAD primitives,are used to extract the features of CAD primitives respectively,and then these features are fused to improve the recognition rate of CAD primitives.Finally,the random forest algorithm is used to identify CAD primitives,and the particle swarm algorithm is used to optimize the random forest algorithm.By designing the inertia weight function of the particle swarm and the dynamic value function of the individual learning factor and the group learning factor,the key parameters of the random forest classification algorithm are optimized.On 1393 primitives,the accuracy rate is 98.78%,which verifies the effectiveness and feasibility of this algorithm.