Past Students¶
The following are postgraduate students who have graduated under my supervision or co-supervision.
Project Students¶
| Student | Degree | Project Title | Abstract |
|---|---|---|---|
 Francois Nelffnelacc@gmail.com | BSc Hons (Computer Science) | Driver Activity Recognition Through Deep Learning | Distracted drivers contribute to a significant proportion of road accidents all over the world. Activities such as texting on cellphones, eating, and reaching for something at the back of the vehicle lead to drivers not paying attention to the road and may result in traffic accidents. This paper proposes the use of residual neural networks (ResNet) with spatio-temporal three-dimensional (3D) kernels to perform distracted driver behaviour recognition. Recently, convolutional neural networks (CNN) with 3D kernels have become an effective tool for action recognition. The 3D kernels extract spatio-temporal features from videos to perform tasks such as human activity recognition. ResNets are a variant of CNNs that utilise skip-connections to realise the training of very deep networks. The large number of parameters in 3D ResNets exposes the possibility of overfitting. Using a large video dataset is thus essential, as it avoids the occurrence of overfitting. This paper examines how different datasets and network depths influence the performance of 3D ResNets. The results are overwhelmingly positive. The findings present a significant positive correlation between the accuracy of a model and the network depth. Furthermore, the quality of the dataset greatly determines the model's ability to generalise effectively. |
 Matisse Ghesquiere matisse.ghesquiere@gmail.com | BSc Hons (Computer Science) | Deep Learning for Plant Disease Detection | In today’s world, plant diseases are a major threat to agriculture crops and their production rate. These are difficult to spot in early stages and it’s not feasible to inspect every leaf manually. During the last decade, computer vision and deep learning models have made a significant progression in classifying various images. We tested different convolutional neural networks on their ability to classify plant diseases. The best model reaches an accuracy of 99.70%. This is based on standardized images, so in order to improve the accuracy for smartphone taken images, more data variety is needed.In general, detecting plant diseases using deep learning models is an excellent approach and much more practical than detection with the human eye. |
Masters Students¶
| Student | Degree | Thesis Title | Abstract |
|---|---|---|---|
 | MSc 2020 University of Fort Hare | Developing a Machine Learning Algorithm for Outdoor Scene Image Segmentation | Image segmentation is one of the major problems in image processing, computer vision and machine learning fields. The main reason for image segmentation existence is to reduce the gap between computer vision and human vision by training computers with different data. Outdoor image segmentation and classification has become very important in the field of computer vision with its applications in woodland-surveillance, defence and security. The task of assigning an input image to one class from a fixed set of categories seem to be a major problem in image segmentation. The main question that has been addressed in this research is how outdoor image classification algorithms can be improved using Region-based Convolutional Neural Network (R-CNN) architecture. There has been no one segmentation method that works best on any given problem. To determine the best segmentation method for a certain dataset, various tests have to be done in order to achieve the best performance. However deep learning models have often achieved increasing success due to the availability of massive datasets and the expanding model depth and parameterisation. In this research Convolutional Neural Network architecture is used in trying to improve the implementation of outdoor scene image segmentation algorithms, empirical research method was used to answer questions about existing image segmentation algorithms and the techniques used to achieve the best performance. Outdoor scene images were trained on a pre-trained region-based convolutional neural network with Visual Geometric Group-16 (VGG-16) architecture. A pre-trained R-CNN model was retrained on five different sample data, the samples had different sizes. Sample size increased from sample one to five, to increase the size on the last two samples the data was duplicated. 21 test images were used to evaluate all the models. Researchers has shown that deep learning methods perform better in image segmentation because of the increase and availability of datasets. The duplication of images did not yield the best results; however, the model performed well on the first three samples. |