Intelligent Fruit Inspection System: Developing a YOLO-based Model for Identifying Defects on Plums Surface.
This research focuses on developing an AI solution using the YOLOv5 algorithm to efficiently sort good and damaged African pears, which are vital crops for African farmers. The model, trained on a dataset of about 2900 African pear images, achieved a mean average precision of 85.1%. By automating the process of detecting and removing defective fruit, this technology has the potential to increase efficiency, reduce waste, improve crop quality, and enhance farm incomes for African pear farmers
Towards an empathic driving assistant: Explainable emotion recognition from driving context
This thesis focuses on driver emotion recognition based on driving context, using techniques like Skope Rules or Decision Trees to predict emotions and infer possible reasons for them. The approach outperforms prior work with a higher AUC-ROC score (63.13% compared to 52.32%) and provides a valuable means of communicating the extracted reasons to drivers for feedback and model improvement.
Selected Medicinal Plants Leaves Identification: A Computer Vision Approach
In this thesis, we investigate different convolutional neural networks and machine learning algorithms for identifying and classifying leaves of three species of the genus Brachylaena. The three species considered are Brachylaena discolor, Brachylaena ilicifolia and Brachylaena elliptica. All three species are used medicinally by people in South Africa to treat diseases like
diabetes. From 1259 labelled images of those plants species (at least 400 for each species) split
into training, evaluation and test sets, we trained and evaluated different deep convolutional
neural networks and machine learning models. The VGG model achieved the best results
with 98.26% accuracy from cross-validation.
Computer Vision for Segmentation and Quantification of Damage Surfaces on African Plum Fruits
The thesis proposes an approach using deep convolutional neural networks with semantic segmentation to quantify the damaged part of the plum fruits, African pears vital crops for African farmers, achieving an average accuracy of 61% on the test set by segmenting and classifying the spoiled plum image pixels.
Runtime configurable deep neural networks for power-efficient adaptive architectures
The thesis explores, using Convolutional Neural Networks, optimizing the training process, and implementing dynamic reconfigurability of the learning algorithm for efficient hardware utilization, aiming to match the allocation of resources to application demands in an efficient manner. It also considers techniques like approximate computing and content adaptivity to achieve power savings without significant loss in accuracy.