Automatic Information Acquisition System Of Incomplete Information Game Situation Based On Deep Learning

Since the emergence of the concept of artificial intelligence,human beings have never stopped exploring the machine game.Machine game can be divided into complete information game and incomplete information game according to its understanding of game situation information.For example,go,chess and other games are complete information games,mahjong and bridge are incomplete information games.based on the problem of incomplete information computer game,this paper proposes to use deep learning to realize the automatic acquisition of incomplete information game situation information,aiming to automatically obtain the situation information of the game scene and upload it to the incomplete information game intellligent decition system,thus enabling the robot to comlete the game completely autonomously.The main research work includes the following parts:1.This paper designs a method of automatically obtaining the situation information of incomplete information game based on template.According to the specified incomplete information game platform,measure its distribution area,segment the situation instance through the traditional image processing methods such as grayscale,binarization,morphological processing and Canny operator edge detection,then combine with convolution neural network to identify the hand information in the game,through the platform template and traditional image processing methods as well as deep learning method In this paper,we use the template method to extract the situation information of incomplete information game.In the process of implementation,this paper contains more than 17000 hand recognition data sets.The experimental results show that the template based method can achieve accurate situation information extraction for the specified platform.In order to adapt this method to different platforms,a parameterized template segmentation method is proposed.The parameters determined during the segmentation of a situation instance are extracted into a set of segmentation parameters with generalization features,forming a unified segmentation framework.2.This paper designs an automatic method to obtain situation information of incomplete information game based on deep learning.This method applies the mask designed by he Kaiming's team R-cnn,through resnet101 and feature pyramid network,extracts the features of the game situation image,then uses the region feature extraction network to identify the opponent card region,finally obtains the semantic segmentation result through neural network to complete the case segmentation,realizes the end-to-end image to case segmentation,while ensuring the performance,solves the template method generalization ability Poor,the process is tedious,and can only be applied to the designated platform.In the process of design and implementation,the incomplete information game situation information data set is constructed,and 1000 real game situations of each platform are marked in detail in the data set.The experimental results show that although the accuracy of the method based on deep learning is slightly worse than that based on template,it is still in the usable range.On the whole,the method based on deep learning is better than that based on template.3.An automatic situation information acquisition system of incomplete information game based on deep learning is designed.This system takes Android based applications as the front end,and takes the server of the flash framework developed based on Python language as the back end.The application is suspended in a game scenario in the form of a suspension ball.The user clicks the suspension ball to intercept the current scene information and upload it to the server,and the information is extracted by the deep learning model.The main purpose of the system is to extract the game situation information of players from the situation image and provide complete game situation information for the intelligent decision-making of incomplete information game.