Comparative analysis of YOLOV5 and YOLOV8 for automated fish detection and classification in underwater environments
- Authors: Kuhlane, Luxolo
- Date: 2024-10-11
- Subjects: Artificial intelligence , Deep learning (Machine learning) , Machine learning , Neural networks (Computer science) , You Only Look Once , YOLOv5 , YOLOv8
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/464333 , vital:76502
- Description: The application of traditional manual techniques for fish detection and classification faces significant challenges, primarily stemming from their labour-intensive nature and limited scalability. Automating these kinds of processes through computer vision practices and machine learning techniques has emerged as a potential solution in recent years. With the development of and increase in ease of access to new technology in recent years, the use of a deep learning object detector known as YOLO (You Only Look Once) in the detection and classification of fish has steadily become notably popular. This thesis thus explores suitable YOLO architectures for detecting and classifying fish. The YOLOv5 and YOLOv8 models were evaluated explicitly for detecting and classifying fish in underwater environments. The selection of these models was based on a literature review highlighting their success in similar applications but remains largely understudied in underwater environments. Therefore, the effectiveness of these models was evaluated through comprehensive experimentation on collected and publicly available underwater fish datasets. In collaboration with the South African Institute of Biodiversity (SAIAB), five datasets were collected and manually annotated for labels for supervised machine learning. Moreover, two publicly available datasets were sourced for comparison to the literature. Furthermore, after determining that the smallest YOLO architectures are better suited to these imbalanced datasets, hyperparameter tuning tailored the models to the characteristics of the various underwater environments used in the research. The popular DeepFish dataset was evaluated to establish a baseline and feasibility of these models in the understudied domain. The results demonstrated high detection accuracy for both YOLOv5 and YOLOv8. However, YOLOv8 outperformed YOLOv5, achieving 97.43% accuracy compared to 94.53%. After experiments on seven datasets, trends revealed YOLOv8’s enhanced generalisation accuracy due to architectural improvements, particularly in detecting smaller fish. Overall, YOLOv8 demonstrated that it is the better fish detection and classification model on diverse data. , Thesis (MSc) -- Faculty of Science, Computer Science, 2024
- Full Text:
Statistical classification, an application to credit default
- Authors: Sikhakhane, Anele Gcina
- Date: 2024-10-11
- Subjects: Binary classification , Default (Finance) , Credit cards , Credit risk , Machine learning , Variables (Mathematics)
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/465069 , vital:76570
- Description: Statistical learning has been used in both industry and academia to create credit scoring models. These models are used to predict who might default on their loan repayments, thus minimizing the risk financial institutions face. In this study six traditional and one more recent classifier, namely kNN, LDA, CART, RF, AdaBoost, XGBoost and SynBoost were used to predict who might default on their loans. The data set used in this study was imbalanced thus sampling and performance evaluation techniques were investigated and used to balance the class distribution and assess the classifiers performance. In addition to the standard variables and data set, new variables called synthetic variables and synthetic data sets were produced, investigated and used to predict who might default on their loans. This study found that the synthetic data set had strong predictive power and sampling methods negatively affected the classifiers performance. The best-performing classifier was XGBoost, with an AUC score of 0.7732. , Thesis (MSc) -- Faculty of Science, Statistics, 2024
- Full Text:
Suspicious activity reports: Enhancing the detection of terrorist financing and suspicious transactions in migrant remittances
- Authors: Mbiva, Stanley Munamato
- Date: 2024-10-11
- Subjects: Migrant remittances , Terrorism financing , Machine learning , Outliers (Statistics) , Anomaly detection (Computer security) , Unsupervised learning
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/465058 , vital:76569
- Description: Migrant remittances have become an important factor in poverty alleviation and microeconomic development in low-income nations. Global migrant remittances are expected to exceed US $630 billion by 2023, according to the World Bank. In addition to offering an alternate source of income that supplements the recipient’s household earnings, they are less likely to be affected by global economic downturns, ensuring stability and a consistent stream of revenue. However, the ease of global migrant remittance financial transfers has attracted the risk of being abused by terrorist organizations to quickly move and conceal operating cash, hence facilitating terrorist financing. This study aims to develop an unsupervised machine-learning model capable of detecting suspicious financial transactions associated with terrorist financing in migrant remittances. The data used in this study came from a World Bank survey of migrant remitters in Belgium. To understand the natural structures and grouping in the dataset, agglomerative hierarchical clustering and k-prototype clustering techniques were employed. This established the number of clusters present in the dataset making it possible to compare individual migrant remittances in the dataset with their peers. A Structural Equation Model (SEM) and an Local Outlier Factor - Isolation Forest (LOF-IF) algorithm were applied to analyze and detect suspicious transactions in the dataset. A traditional Rule-Based Method (RBM) was also created as a benchmark algorithm that evaluates model performance. The results show that the SEM model classifies a significantly high number of transactions as suspicious, making it prone to detecting false positives. Finally, the study applied the proposed ensemble outlier detection model to detect suspicious transactions in the same data set. The proposed ensemble model utilized an Isolation Forest (IF) for pruning and a Local Outlier Factor (LOF) to detect local outliers. The model performed exceptionally well, being able to detect over 90% of suspicious transactions in the testing data set during model cross-validation. , Thesis (MSc) -- Faculty of Science, Statistics, 2024
- Full Text:
Towards an artificial intelligence-based agent for characterising the organisation of primes
- Authors: Oyetunji, Nicole Armlade
- Date: 2024-04-04
- Subjects: Numbers, Prime , Odd number , Machine learning , Deep learning (Machine learning) , Mathematical forecasting , Neural networks (Computer science) , Artificial intelligence
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/435389 , vital:73153
- Description: Machine learning has experienced significant growth in recent decades, driven by advancements in computational power and data storage. One of the applications of machine learning is in the field of number theory. Prime numbers hold significant importance in mathematics and its applications, for example in cryptography, owing to their distinct properties. Therefore, it is crucial to efficiently obtain the complete list of primes below a given threshold, with low relatively computational cost. This study extensively explores a deterministic scheme, proposed by Hawing and Okouma (2016), that is centered around Consecutive Composite Odd Numbers, showing the link between these numbers and prime numbers by examining their internal structure. The main objective of this dissertation is to develop two main artificial intelligence agents capable of learning and recognizing patterns within a list of consecutive composite odd numbers. To achieve this, the mathematical foundations of the deterministic scheme are used to generate a dataset of consecutive composite odd numbers. This dataset is further transformed into a dataset of differences to simplify the prediction problem. A literature review is conducted which encompasses research from the domains of machine learning and deep learning. Two main machine learning algorithms are implemented along with their variations, Long Short-Term Memory Networks and Error Correction Neural Networks. These models are trained independently on two separate but related datasets, the dataset of consecutive composite odd numbers and the dataset of differences between those numbers. The evaluation of these models includes relevant metrics, for example, Root Mean Square Error, Mean Absolute Percentage Error, Theil U coefficient, and Directional Accuracy. Through a comparative analysis, the study identifies the top-performing 3 models, with a particular emphasis on accuracy and computational efficiency. The results indicate that the LSTM model, when trained on difference data and coupled with exponential smoothing, displays superior performance as the most accurate model overall. It achieves a RMSE of 0.08, which significantly outperforms the dataset’s standard deviation of 0.42. This model exceeds the performance of basic estimator models, implying that a data-driven approach utilizing machine learning techniques can provide valuable insights in the field of number theory. The second best model, the ECNN trained on difference data combined with exponential smoothing, achieves an RMSE of 0.28. However, it is worth mentioning that this model is the most computationally efficient, being 32 times faster than the LSTM model. , Thesis (MSc) -- Faculty of Science, Mathematics, 2024
- Full Text:
Natural Language Processing with machine learning for anomaly detection on system call logs
- Authors: Goosen, Christo
- Date: 2023-10-13
- Subjects: Natural language processing (Computer science) , Machine learning , Information security , Anomaly detection (Computer security) , Host-based intrusion detection system
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/424699 , vital:72176
- Description: Host intrusion detection systems and machine learning have been studied for many years especially on datasets like KDD99. Current research and systems are focused on low training and processing complex problems such as system call returns, which lack the system call arguments and potential traces of exploits run against a system. With respect to malware and vulnerabilities, signatures are relied upon, and the potential for natural language processing of the resulting logs and system call traces needs further experimentation. This research looks at unstructured raw system call traces from x86_64 bit GNU Linux operating systems with natural language processing and supervised and unsupervised machine learning techniques to identify current and unseen threats. The research explores whether these tools are within the skill set of information security professionals, or require data science professionals. The research makes use of an academic and modern system call dataset from Leipzig University and applies two machine learning models based on decision trees. Random Forest as the supervised algorithm is compared to the unsupervised Isolation Forest algorithm for this research, with each experiment repeated after hyper-parameter tuning. The research finds conclusive evidence that the Isolation Forest Tree algorithm is effective, when paired with a Principal Component Analysis, in identifying anomalies in the modern Leipzig Intrusion Detection Data Set (LID-DS) dataset combined with samples of executed malware from the Virus Total Academic dataset. The base or default model parameters produce sub-optimal results, whereas using a hyper-parameter tuning technique increases the accuracy to within promising levels for anomaly and potential zero day detection. , Thesis (MSc) -- Faculty of Science, Computer Science, 2023
- Full Text:
Selected medicinal plants leaves identification: a computer vision approach
- Authors: Deyi, Avuya
- Date: 2023-10-13
- Subjects: Deep learning (Machine learning) , Machine learning , Convolutional neural network , Computer vision in medicine , Medicinal plants
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/424552 , vital:72163
- Description: Identifying and classifying medicinal plants are valuable and essential skills during drug manufacturing because several active pharmaceutical ingredients (API) are sourced from medicinal plants. For many years, identifying and classifying medicinal plants have been exclusively done by experts in the domain, such as botanists, and herbarium curators. Recently, powerful computer vision technologies, using machine learning and deep convolutional neural networks, have been developed for classifying or identifying objects on images. A convolutional neural network is a deep learning architecture that outperforms previous advanced approaches in image classification and object detection based on its efficient features extraction on images. 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. , Thesis (MSc) -- Faculty of Science, Mathematics, 2023
- Full Text:
A systematic methodology to evaluating optimised machine learning based network intrusion detection systems
- Authors: Chindove, Hatitye Ethridge
- Date: 2022-10-14
- Subjects: Intrusion detection systems (Computer security) , Machine learning , Computer networks Security measures , Principal components analysis
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/362774 , vital:65361
- Description: A network intrusion detection system (NIDS) is essential for mitigating computer network attacks in various scenarios. However, the increasing complexity of computer networks and attacks makes classifying unseen or novel network traffic challenging. Supervised machine learning techniques (ML) used in a NIDS can be affected by different scenarios. Thus, dataset recency, size, and applicability are essential factors when selecting and tuning a machine learning classifier. This thesis explores developing and optimising several supervised ML algorithms with relatively new datasets constructed to depict real-world scenarios. The methodology includes empirical analyses of systematic ML-based NIDS for a near real-world network system to improve intrusion detection. The thesis is experimental heavy for model assessment. Data preparation methods are explored, followed by feature engineering techniques. The model evaluation process involves three experiments testing against a validation, un-trained, and retrained set. They compare several traditional machine learning and deep learning classifiers to identify the best NIDS model. Results show that the focus on feature scaling, feature selection methods and ML algo- rithm hyper-parameter tuning per model is an essential optimisation component. Distance based ML algorithm performed much better with quantile transformation whilst the tree based algorithms performed better without scaling. Permutation importance performs as a feature selection method compared to feature extraction using Principal Component Analysis (PCA) when applied against all ML algorithms explored. Random forests, Sup- port Vector Machines and recurrent neural networks consistently achieved the best results with high macro f1-score results of 90% 81% and 73% for the CICIDS 2017 dataset; and 72% 68% and 73% against the CICIDS 2018 dataset. , Thesis (MSc) -- Faculty of Science, Computer Science, 2022
- Full Text:
A multispectral and machine learning approach to early stress classification in plants
- Authors: Poole, Louise Carmen
- Date: 2022-04-06
- Subjects: Machine learning , Neural networks (Computer science) , Multispectral imaging , Image processing , Plant stress detection
- Language: English
- Type: Master's thesis , text
- Identifier: http://hdl.handle.net/10962/232410 , vital:49989
- Description: Crop loss and failure can impact both a country’s economy and food security, often to devastating effects. As such, the importance of successfully detecting plant stresses early in their development is essential to minimize spread and damage to crop production. Identification of the stress and the stress-causing agent is the most critical and challenging step in plant and crop protection. With the development of and increase in ease of access to new equipment and technology in recent years, the use of spectroscopy in the early detection of plant diseases has become notably popular. This thesis narrows down the most suitable multispectral imaging techniques and machine learning algorithms for early stress detection. Datasets were collected of visible images and multispectral images. Dehydration was selected as the plant stress type for the main experiments, and data was collected from six plant species typically used in agriculture. Key contributions of this thesis include multispectral and visible datasets showing plant dehydration as well as a separate preliminary dataset on plant disease. Promising results on dehydration showed statistically significant accuracy improvements in the multispectral imaging compared to visible imaging for early stress detection, with multispectral input obtaining a 92.50% accuracy over visible input’s 77.50% on general plant species. The system was effective at stress detection on known plant species, with multispectral imaging introducing greater improvement to early stress detection than advanced stress detection. Furthermore, strong species discrimination was achieved when exclusively testing either early or advanced dehydration against healthy species. , Thesis (MSc) -- Faculty of Science, Ichthyology & Fisheries Sciences, 2022
- Full Text:
Statistical and Mathematical Learning: an application to fraud detection and prevention
- Authors: Hamlomo, Sisipho
- Date: 2022-04-06
- Subjects: Credit card fraud , Bootstrap (Statistics) , Support vector machines , Neural networks (Computer science) , Decision trees , Machine learning , Cross-validation , Imbalanced data
- Language: English
- Type: Master's thesis , text
- Identifier: http://hdl.handle.net/10962/233795 , vital:50128
- Description: Credit card fraud is an ever-growing problem. There has been a rapid increase in the rate of fraudulent activities in recent years resulting in a considerable loss to several organizations, companies, and government agencies. Many researchers have focused on detecting fraudulent behaviours early using advanced machine learning techniques. However, credit card fraud detection is not a straightforward task since fraudulent behaviours usually differ for each attempt and the dataset is highly imbalanced, that is, the frequency of non-fraudulent cases outnumbers the frequency of fraudulent cases. In the case of the European credit card dataset, we have a ratio of approximately one fraudulent case to five hundred and seventy-eight non-fraudulent cases. Different methods were implemented to overcome this problem, namely random undersampling, one-sided sampling, SMOTE combined with Tomek links and parameter tuning. Predictive classifiers, namely logistic regression, decision trees, k-nearest neighbour, support vector machine and multilayer perceptrons, are applied to predict if a transaction is fraudulent or non-fraudulent. The model's performance is evaluated based on recall, precision, F1-score, the area under receiver operating characteristics curve, geometric mean and Matthew correlation coefficient. The results showed that the logistic regression classifier performed better than other classifiers except when the dataset was oversampled. , Thesis (MSc) -- Faculty of Science, Statistics, 2022
- Full Text:
Application of machine learning, molecular modelling and structural data mining against antiretroviral drug resistance in HIV-1
- Authors: Sheik Amamuddy, Olivier Serge André
- Date: 2020
- Subjects: Machine learning , Molecules -- Models , Data mining , Neural networks (Computer science) , Antiretroviral agents , Protease inhibitors , Drug resistance , Multidrug resistance , Molecular dynamics , Renin-angiotensin system , HIV (Viruses) -- South Africa , HIV (Viruses) -- Social aspects -- South Africa , South African Natural Compounds Database
- Language: English
- Type: text , Thesis , Doctoral , PhD
- Identifier: http://hdl.handle.net/10962/115964 , vital:34282
- Description: Millions are affected with the Human Immunodeficiency Virus (HIV) world wide, even though the death toll is on the decline. Antiretrovirals (ARVs), more specifically protease inhibitors have shown tremendous success since their introduction into therapy since the mid 1990’s by slowing down progression to the Acquired Immune Deficiency Syndrome (AIDS). However, Drug Resistance Mutations (DRMs) are constantly selected for due to viral adaptation, making drugs less effective over time. The current challenge is to manage the infection optimally with a limited set of drugs, with differing associated levels of toxicities in the face of a virus that (1) exists as a quasispecies, (2) may transmit acquired DRMs to drug-naive individuals and (3) that can manifest class-wide resistance due to similarities in design. The presence of latent reservoirs, unawareness of infection status, education and various socio-economic factors make the problem even more complex. Adequate timing and choice of drug prescription together with treatment adherence are very important as drug toxicities, drug failure and sub-optimal treatment regimens leave room for further development of drug resistance. While CD4 cell count and the determination of viral load from patients in resource-limited settings are very helpful to track how well a patient’s immune system is able to keep the virus in check, they can be lengthy in determining whether an ARV is effective. Phenosense assay kits answer this problem using viruses engineered to contain the patient sequences and evaluating their growth in the presence of different ARVs, but this can be expensive and too involved for routine checks. As a cheaper and faster alternative, genotypic assays provide similar information from HIV pol sequences obtained from blood samples, inferring ARV efficacy on the basis of drug resistance mutation patterns. However, these are inherently complex and the various methods of in silico prediction, such as Geno2pheno, REGA and Stanford HIVdb do not always agree in every case, even though this gap decreases as the list of resistance mutations is updated. A major gap in HIV treatment is that the information used for predicting drug resistance is mainly computed from data containing an overwhelming majority of B subtype HIV, when these only comprise about 12% of the worldwide HIV infections. In addition to growing evidence that drug resistance is subtype-related, it is intuitive to hypothesize that as subtyping is a phylogenetic classification, the more divergent a subtype is from the strains used in training prediction models, the less their resistance profiles would correlate. For the aforementioned reasons, we used a multi-faceted approach to attack the virus in multiple ways. This research aimed to (1) improve resistance prediction methods by focusing solely on the available subtype, (2) mine structural information pertaining to resistance in order to find any exploitable weak points and increase knowledge of the mechanistic processes of drug resistance in HIV protease. Finally, (3) we screen for protease inhibitors amongst a database of natural compounds [the South African natural compound database (SANCDB)] to find molecules or molecular properties usable to come up with improved inhibition against the drug target. In this work, structural information was mined using the Anisotropic Network Model, Dynamics Cross-Correlation, Perturbation Response Scanning, residue contact network analysis and the radius of gyration. These methods failed to give any resistance-associated patterns in terms of natural movement, internal correlated motions, residue perturbation response, relational behaviour and global compaction respectively. Applications of drug docking, homology-modelling and energy minimization for generating features suitable for machine-learning were not very promising, and rather suggest that the value of binding energies by themselves from Vina may not be very reliable quantitatively. All these failures lead to a refinement that resulted in a highly sensitive statistically-guided network construction and analysis, which leads to key findings in the early dynamics associated with resistance across all PI drugs. The latter experiment unravelled a conserved lateral expansion motion occurring at the flap elbows, and an associated contraction that drives the base of the dimerization domain towards the catalytic site’s floor in the case of drug resistance. Interestingly, we found that despite the conserved movement, bond angles were degenerate. Alongside, 16 Artificial Neural Network models were optimised for HIV proteases and reverse transcriptase inhibitors, with performances on par with Stanford HIVdb. Finally, we prioritised 9 compounds with potential protease inhibitory activity using virtual screening and molecular dynamics (MD) to additionally suggest a promising modification to one of the compounds. This yielded another molecule inhibiting equally well both opened and closed receptor target conformations, whereby each of the compounds had been selected against an array of multi-drug-resistant receptor variants. While a main hurdle was a lack of non-B subtype data, our findings, especially from the statistically-guided network analysis, may extrapolate to a certain extent to them as the level of conservation was very high within subtype B, despite all the present variations. This network construction method lays down a sensitive approach for analysing a pair of alternate phenotypes for which complex patterns prevail, given a sufficient number of experimental units. During the course of research a weighted contact mapping tool was developed to compare renin-angiotensinogen variants and packaged as part of the MD-TASK tool suite. Finally the functionality, compatibility and performance of the MODE-TASK tool were evaluated and confirmed for both Python2.7.x and Python3.x, for the analysis of normals modes from single protein structures and essential modes from MD trajectories. These techniques and tools collectively add onto the conventional means of MD analysis.
- Full Text:
Technology in conservation: towards a system for in-field drone detection of invasive vegetation
- Authors: James, Katherine Margaret Frances
- Date: 2020
- Subjects: Drone aircraft in remote sensing , Neural networks (Computer science) , Drone aircraft in remote sensing -- Case studies , Machine learning , Computer vision , Environmental monitoring -- Remote sensing , Invasive plants -- Monitoring
- Language: English
- Type: text , Thesis , Masters , MSc
- Identifier: http://hdl.handle.net/10962/143408 , vital:38244
- Description: Remote sensing can assist in monitoring the spread of invasive vegetation. The adoption of camera-carrying unmanned aerial vehicles, commonly referred to as drones, as remote sensing tools has yielded images of higher spatial resolution than traditional techniques. Drones also have the potential to interact with the environment through the delivery of bio-control or herbicide, as seen with their adoption in precision agriculture. Unlike in agricultural applications, however, invasive plants do not have a predictable position relative to each other within the environment. To facilitate the adoption of drones as an environmental monitoring and management tool, drones need to be able to intelligently distinguish between invasive and non-invasive vegetation on the fly. In this thesis, we present the augmentation of a commercially available drone with a deep machine learning model to investigate the viability of differentiating between an invasive shrub and other vegetation. As a case study, this was applied to the shrub genus Hakea, originating in Australia and invasive in several countries including South Africa. However, for this research, the methodology is important, rather than the chosen target plant. A dataset was collected using the available drone and manually annotated to facilitate the supervised training of the model. Two approaches were explored, namely, classification and semantic segmentation. For each of these, several models were trained and evaluated to find the optimal one. The chosen model was then interfaced with the drone via an Android application on a mobile device and its performance was preliminarily evaluated in the field. Based on these findings, refinements were made and thereafter a thorough field evaluation was performed to determine the best conditions for model operation. Results from the classification task show that deep learning models are capable of distinguishing between target and other shrubs in ideal candidate windows. However, classification in this manner is restricted by the proposal of such candidate windows. End-to-end image segmentation using deep learning overcomes this problem, classifying the image in a pixel-wise manner. Furthermore, the use of appropriate loss functions was found to improve model performance. Field tests show that illumination and shadow pose challenges to the model, but that good recall can be achieved when the conditions are ideal. False positive detection remains an issue that could be improved. This approach shows the potential for drones as an environmental monitoring and management tool when coupled with deep machine learning techniques and outlines potential problems that may be encountered.
- Full Text: