Study On The Key Technologies Of The Cigarettes Recognition

This thesis discusses the use of the Mask R-CNN convolutional neural network based on the detection model(developed by Facebook AI Research)for obj ect recognition.Training was conducted on dataset(database)of images that contain classes of objects of illegal nature.The thesis contains a description of the stages of training a neural network and testing a trained model.Also provided is a summary table of test results of a trained model,containing the accuracy and number of false positives of the neural network for each class.Testing of the model before and after training was carried out for a comparative analysis of the learning ability of the neural network for new classes of objects without impairing the accuracy for existing classes.The results of the comparison of the proposed models are presented in the summary table.Challenges of accuracy and loss in Mask R-CNN are highlighted in the context.The state-of-the-art of object recognition is shown with detailed explanation.Easy generalization to other tasks,e.g.,allowing us to estimate smoked cigarettes in the same framework.