Refine
Has Fulltext
- yes (2)
Year of publication
- 2019 (2) (remove)
Document Type
- Master's Thesis (2)
Language
- English (2) (remove)
Is part of the Bibliography
- yes (2)
Keywords
- Neural Networks (2) (remove)
Institute
Search Engine Opt imisation, also known as SEO, is one of the online marketing channels t hat, when it is s et u p s uitably, it could continue to pay dividends over time without investment. Recently, SEO teams of some companies keep investigating historical data to predict the future trends of revenue associated with the number of clicks, number of imp ressions, and number of sear ches which is prop osed to help with companies’ quality planning and campaign investment. Some of the challenges experienced by SEO analysts when attempting to forecast the revenue is that there is currently no way to standardize or forecast customers’ behaviour, which means the trends could be different every day, month, and year. In this research, SEO traffic data from one of the online travel a gencies are collected for the purpose of data exploration, analysis and forecasting which are expected to bring business values and give some be neficialinsights. Moreover, different time-series forecasting models are selected to conduct experiments seeking the best fit model for SEO data; Autoregressive Integrated Moving Average (ARIMA) model is initially performed, followed by Long-Short Term Memory (LSTM) of Recurrent Neural Networks (RNNs). As a result, it is proved that ARIMA is yet a classical statistics model but powerful enough for such small-size data, albeit the data is non-stationary and has too much white noise. Me anwhile, the LSTM is a deep learning tool which could deal with different types of data, but still need to be applied with a larger size of data to prove its competence.
The goal of this research is to investigate the use of deep convolutional neural networks for racing bib number recognition in sport images. Several deep neural network architectures are studied. Three final architectures are trained on three different sets of data: 1) Street View House Numbers (SVHN) Dataset, 2) A private dataset from Flashframe.io from different running events, and 3) A combination of dataset 1 and 2. This thesis investigates the performance that can be obtained on racing bib numbers from a neural network that has been trained on solely images from street house numbers, on a mixture of SVHN and RBN images as well as only on RBN images. The motivation behind this is to see how well this problem can be solved by transfer learning, as labelled images of racing bib numbers are scarce.
The models are tested on the RBNR Dataset (Ben-Ami et al., 2012) and a subset of the private dataset from Flashframe.io. The study shows that the best recognition results were obtained by a model trained on the hybrid dataset of all SVHN images plus an additional 50.000 images. This model outperformed the models that had been trained solely on the SVHN Dataset or the private racing bib number dataset.
The best model resulted in Recall of 0,92, Precision of 0,93 and F-measure of 0,93 on the RBNR Dataset (using the same formulas as previously reported on the RBNR dataset), and 0,97, 0,97 and 0,97 on the private dataset, respectively. The reported recognition results on the RBNR dataset are much higher than previously used methods and proves that neural networks can effectively be used for racing bib number recognition.