Sport News Classification with Convolutional Neural Network and Long-Short Term Memory


In this work, a system has been designed and implemented to classify the sports types found in sports news videos with Long-Short Term Memory Network (LSTM) method and to automatically segment the sport news program. Sports news we encounter in daily life contains anchor part who present news. There is no need to classify these parts. Thus, anchor faces was detected with using Local Binary Pattern Histogram method and removed from sport news in preprocessing step. In order to increase the processing speed instead of performing face recognition to the entire video, scene transitions were determined by using 64-bin histogram method and k-means algorithm and face recognition was performed only at this scene transitions. After the anchors are removed from videos, CNNs were used for feature extraction and LSTM was used for classification of videos by sport. In order to train and test the system, UCF-101 sport dataset, Unsegmented Sport News data set and YouTube sport videos dataset which is created from sport videos from Youtube by us were used. Classification was performed for 9 sport types including swimming, American football, football, tennis, basketball, wrestling, baseball, volleyball, and golf.



Machine learning algorithms

Software And Hardware

• Hardware: Processor: i3 ,i5 RAM: 4GB Hard disk: 16 GB • Software: operating System : Windws2000/XP/7/8/10 Anaconda,jupyter,spyder,flask Frontend :-python Backend:- MYSQL