Research And Application Of Incomplete Information Game Situation Data Acquisition Based On Mask RCNN

As a common problem in daily life,machine game is attracting more and more attention.Many researchers try to study the machine game in various complex backgrounds,especially the non-perfect information game.Early researchers usually need to manually obtain and process a large number of data,the process is tedious and tedious,which brings a lot of inconvenience to the research.Then people realized the importance of automatic access to information,but much of the data is intuitive to humans and abstract to computers.With the development of artificial intelligence industry,people begin to use the method of deep learning to study this problem.Study the complete information game is the most important subject of situation information,both on the one hand the decision results,and can help in decision-making,based on the line as the research object of the complete information game of mahjong games,through the analysis of the graphical interface,using image processing method is proposed for information on the situation.The main research work completed is as follows:1.Designed a method to obtain incomplete information game situation data based on the Mask RCNN instance segmentation algorithm,extracted features by combining with the ResNet101 feature pyramid network,extracted the network to identify the information in the target area by using regional features,and obtained semantic segmentation results through the neural network to achieve instance segmentation.Compared with the recognition results of Yolo v3 model,the experimental results show that:the recognition accuracy of Mask RCNN instance segmentation is higher and the detection speed is slower,but it is more suitable for the game situation data of incomplete information that needs to be accurately acquired.2.Three methods were designed to transplant the Mask RCNN model to the iOS mobile terminal.Considering the development of mobile applications and mobile devices is higher and higher performance,at the same time in order to solve the network transmission,the utilization of resource and user privacy issues,this paper puts forward three methods will be trained instance segmentation model into iOS mobile terminal,the experimental results show that:CoreML framework does not apply to convert some deep learning model;The way to use the Python interpreter is not conducive to practical use due to the excessive capacity of the interpreter;The Mask RCNN model trained by Tensorflow object detection API was successfully transplanted with OpenCV to realize the segmentation and recognition of incomplete information game situation data on the mobile terminal.