This research, titled "Shot-Net: A Convolutional Neural Network for Classifying Different Shots in Cricket," submitted by Md. Ferdouse Ahmed Foysal and Mohammad Shakirul Islam to the Department of Computer Science and Engineering, Faculty of Science and Information Technology, Daffodil International University, has been accepted as satisfactory in partial fulfillment of the requirements for the degree of Bachelor of Engineering in Computer Science and Computing and endorsed for style and content. Department of Computer Science and Engineering Faculty of Science and Information Technology Daffodil International University.
We hereby declare that this research was conducted by us under the supervision of Mr. Nafis Neehal, Lecturer, Department of Computer Science and Engineering and co-supervision of Enamul Karim, Lecturer, Department of Computer Science and Engineering, Faculty of Science and Information. Technology, Daffodil International University. Department of Computer Science and Engineering Faculty of Science and Information Technology Daffodil International University.
Enamul Karim, Lecturer, Department of CSE, Faculty of Science and Information Technology, Daffodil International University, Dhaka. Syed Akhter Hossain, Professor and Head, Department of Computer Science and Engineering, Faculty of Science and Information Technology, DIU, for his valuable support and advice to complete our project and also heartiest thanks to other faculty members and staff of the Department of CSE, Daffodil International University . Applying computer vision and machine learning techniques in cricket for various analysis is an emerging domain now.
Therefore, we propose a novel approach to classify different types of cricket shots using Convolutional Neural Network and Deep Learning.
Objectives
Motivation
Rational of the Study
Research Questions
Expected Outcome…..…………………………………………………………... 4-5
Chapter 1 has demonstrated an introduction to the project with objectives, motivation, research questions and expected results. This section describes the entire layout of this report. Then the later section of this second chapter shows the extent of their restriction of this field. In order to discuss the theoretical part of the research, this chapter elaborates on the statistical methods in this work.
And in the last section of this chapter, to validate the model and to show the accuracy mark of the classifier, a confusion matrix analysis is presented. We conclude the chapter by showing the limitations of our work, which may be the future scope of others who wish to work in this area.
BACKGROUND 6-9
Related Works ……………………………………..………………………….. 6-7
CNN means that Convolutional Neural Network is one of the strongest networks in the field of deep learning. The chapter concludes with an explanation of the statistical theories of our project and furthermore with a clear concept of the implementation requirements. The Anaconda application was used for the entire training and testing process. Anaconda is a free and open-source distribution of the Python and R programming languages for data science and machine learning applications.
We took 80% image of the dataset which means 2880 images to train the model and 20% image of the dataset which means 720 images for testing. In the train dataset, each class contains 480 images and in the test dataset, each class contains 120 images. As we don't have better GPU in our computer which we have used to train the model .So we used grayscale images to train the model.
Finally, there are two dense layers, where in one we used 'linear' as activation function and in the other we used 'soft-max' as activation function. It seems like a strange combination of biology and math with a little CS sprinkled in, but these networks have been some of the most influential innovations in the field of computer vision. Depending on the resolution and size of the image, it will see a 32 x 32 x 3 set of numbers (the 3 refers to the RGB values).
I would strongly encourage those interested to read up on them and understand their function and effects, but in a general sense they provide non-linearities and preservation of dimensions that help improve network robustness and control overfitting. In the case of a CNN, convolution is performed on the input data using a filter or a kernel to then produce a feature map. The goal is to downsample an input representation, reducing its dimensionality and allowing assumptions to be made about features contained in the subspaces that are stored.
It is basically used to reduce the size of the image because the larger number of pixels contributes to more parameters that can encompass large amounts of data. Various types of functions such as softmax activation function, SVM and many others are used here for high-level reasoning in the neural network. This means that it can combine all the local features found from the previous convolutional layers.
Each feature map channel in the output of a CNN layer is a "flattened" 2D array created by adding the results of multiple 2D kernels (one for each channel in the input layer). The training dataset has 2880 images, so we used 2304 images for training and 576 images for validation. The overall process of the model is divided into few steps such as dataset collection, data preparation, data augmentation, data resizing, proposed model description and final training procedure of the model.
The two measures are sometimes used together in the F1 score (or f-measure) to provide a single measurement for a system.
