Research Bank
Not a member yet
5195 research outputs found
Sort by
Empowering community and philanthropic funders: The case for investment in non-market housing
No full textAotearoa New Zealand is in the middle of a prolonged housing crisis. This is marked by a chronic undersupply of private and non-market housing which has created social, health, and economic inequities. Māori communities are disproportionately affected, consistently experiencing the highest rates of housing insecurity, overcrowding, and unaffordability. Over 20,000 households remain on the social housing waitlist, while private rental costs continue to escalate. The reasons for the housing crisis are complex, structural and deeply entrenched. Housing access and affordability also remain poor across other developed countries, with particularly severe challenges in Commonwealth nations like Canada and Australia, where rising costs and stagnant wages have made homeownership increasingly unattainable for many.
This research inquiry investigates how philanthropic and community funders can use impact investments to play their part in addressing the housing crisis. A pragmatism mixed-methodology model was used, including desktop research, namely reviewing literature and data analysis, semi-structured interviews and surveys. This employed purposive sampling and the participants were funders, third-party financiers and housing providers to give a sector wide overview. The exploration shows that philanthropic capital can increase non-market housing supply by providing adaptable, long-term funding unavailable through traditional markets. Collaborative financing, integrating public, private, and charitable resources, is essential to increase housing supply. Sector capacity requires strengthening to manage intricate financial and development arrangements. For Māori, there is a need for culturally responsive and flexible funding, particularly on whenua Māori, where standard financing is unavailable due to restrictive banking and regulatory conditions.
Findings advocate for a centralised financing platform to consolidate diverse capital sources and bolster the sector’s financial and governance expertise. Regulatory adjustments are proposed to ease borrowing costs and align frameworks with social housing’s extended, low-profit timelines. Tailored funding solutions for Māori, crafted with iwi, government, and philanthropists, are deemed necessary to address systemic obstacles and advance self-determination. Measuring social outcomes via tools like Social Return on Investment (SROI) is recommended to demonstrate value and draw additional investment. Strong, bipartisan political resolve is urged to allocate substantial resources and coordinate efforts across parties. These measures position philanthropic and community funders as central to expanding non-market housing, offering a practical means to mitigate Aotearoa New Zealand’s housing crisis and foster equitable outcomes, especially for Māori
Humanizing AI chatbots: The role of speech emotion recognition (SER) with deep learning for improved user experience
No full textThis research explores the enhancement of AI chatbots through the integration of Speech Emotion Recognition (SER) capabilities, aiming to create more emotionally intelligent and responsive systems. By using advanced Deep Learning techniques such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), the study seeks to improve the accuracy and robustness of SER models in recognizing and interpreting human emotions from speech. The Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) serves as the primary dataset, providing a various range of emotional speech samples. The study uses comprehensive data preparation and augmentation techniques, including noise injection, speed variation and pitch shifting to simulate real-world conditions and enhance model performance. Key features such as Zero Crossing Rate (ZCR), Chroma, Mel-Frequency Cepstral Coefficients (MFCC), Root Mean Square (RMS) and Mel Spectrogram are extracted to capture the essence of emotional speech.
A detailed literature review sets the stage by summarizing notable contributions and advancements in the field of SER, highlighting the evolution and integration of deep learning methods. The methodological framework outlines the systematic approach to data preparation, feature extraction, model construction and evaluation. Regularization techniques such as Batch Normalization and L2 Regularization are used to prevent overfitting and ensure robust model performance.
The study’s experiments uncover significant improvements in model accuracy, with the best test accuracy reaching 87.5%. This performance is noteworthy when compared to previous studies: Mustaqeem and Kwon achieved an accuracy of approximately 80.2% [1], and Kapoor and Kumar achieved an accuracy of 82.29% [2]. These comparisons highlight the substantial advancement of our model in the field of Speech Emotion Recognition.
The results are visualized through training history plots, demonstrating the model’s learning behaviour and generalization capabilities. The findings highlight the immense potential of SER-enhanced chatbots in various applications, including customer service and mental health support by enabling more empathetic and personalized interactions.
In conclusion, this research advances the field of Speech Emotion Recognition by developing highly accurate and reliable models. The integration of SER capabilities into AI chatbots promises to revolutionize human-technology interactions, making them more meaningful and supportive. This study lays the groundwork for future explorations into emotionally intelligent AI systems, highlighting their potential to significantly enhance user experiences across diverse domains
Image-text-image transmission for accident scene communication: A generative AI approach
No full textRESEARCH QUESTIONS
1. Impact of effective question design on model output accuracy
⚫ How do we design multi-angle questions (e.g., from the perspectives of police, insurance, and news) to ensure that linear models can provide accurate and detailed answers about accident scenes?
⚫ How do different types of question design (e.g., concise vs. complex) affect the quality of generated accident scene descriptions?
2. Selection and deployment of image-to-text and text-to-image models
⚫ In real-time accident communication systems, how do we select and evaluate image-totext and text-to-image models based on criteria such as accuracy, speed, and resource usage?
⚫ How can we improve the real-time accuracy of accident scene analysis by deploying lightweight generative models in edge devices and vehicle cloud systems?
3. Comparison of LLaVA and PixArt-Σ models in accident scene recovery
⚫ How do the text descriptions generated by LLaVA compare with the image recovery effects of the PixArt-Σ model in terms of accuracy, detail richness, and practicality, especially in accident scene understanding?
⚫ How does the quality of the descriptions generated by LLaVA affect the clarity of image recovery and the decision-making effect of drivers in high-speed scenarios?
4. The impact of multi-angle information fusion on model output
⚫ How can we improve the accuracy and richness of the generated accident descriptions by combining multi-angle data (such as news, police reports, and insurance perspectives)?
⚫ How can we optimize the description generation ability of the model in multi-angle information fusion of accident scenes to ensure that the output meets the needs of different users?
5. Impact of network conditions on transmission delay of accident images and text descriptions
⚫ How do different network conditions (e.g., bandwidth, latency) affect the transmission delay of accident images and text descriptions in a high-speed driving environment?
ABSTRACT
The main purpose of this study was to address the challenge of reducing the information transmission delay in accident scenarios during high-speed driving. To achieve this, the study proposes a method for converting accident images into text descriptions for transmission and then re-generating images based on those descriptions. This image-text-image approach optimizes bandwidth usage and reduces emergency response times. By comparing the transmission speeds of text and image data uploaded to the server under simulated similar network conditions, the study shows that text descriptions are significantly better than images in terms of speed and resource efficiency. In addition, the study combines different perspectives - news, insurance reports, and police accident analysis - to enrich the model's understanding of the accident scene and designs and compares the limitations on generating images from different perspectives to help develop an accurate and comprehensive description of the restored scene. The experiments mainly used generative AI models, such as LLaVA and PixArt Sigma, to test the feasibility and quality of information restoration from text to images. The results show that despite some limitations, such as model constraints and input truncation, the quality of images generated based on short questions and descriptions is very similar visually. The proposed method is feasible for improving accident response and communication in bandwidthconstrained environments, highlighting the potential of generative AI in enhancing road safety systems
Spatial analysis of maritime disasters in the Philippines: Distribution patterns and identification of high-risk areas
No full textMaritime accidents frequently occur in the Philippine archipelagic waters, often resulting in significant loss of life. These incidents highlight the urgent need for improvements in the country’s maritime safety systems. By utilising accident data from the Philippine Coast Guard and the GISIS IMO databases, spatial analytical approaches were employed to determine incident distribution patterns and resulted in an overall depiction of the likelihood component of risk across the country’s territorial waters. Kernel density and hotspot analysis revealed areas where incidents were concentrated and where statistically significant hotspots occurred. The Maxent tool was used to develop risk likelihood models for the incident locations using environmental rasters representing wind speed, significant wave height, depth, surface current, land distance and port distance. Model performance metrics including the AUC, TSS and Kappa were used to compare the two datasets and provide confidence on model robustness. Variable contribution figures showed that land distance is the most influential variable, with the majority of high-risk areas predominantly located near population centres. The resulting maps provide an intuitive and informative depiction of the characteristic patterns of maritime accidents in the country, identify areas of high risk requiring immediate attention and offer valuable insights to support strategies for improving and enhancing the country’s maritime safety
Pineapple fields forever. [Decolonising the Filipino identity through screenwriting]
No full textBilingual playscript in English and Tagalog (Filipino
What happens in the shadows: The unknown impact of visitors on captive North Island brown kiwi (Apteryx Mantelli) behaviour
No full textVisitor presence can impact the behaviour of zoo animals, leading to welfare concerns. Studies show some mammals increase vigilance, stereotypies, and activity levels in the presence of high visitor numbers. In contrast to the large number of studies on mammals, birds have been under-represented in the literature on visitor-animal impacts. Our study investigated the behaviour of two indoor kiwis (Apteryx mantelli) housed at Wellington Zoo, Wellington, Aotearoa New Zealand, during school holidays (SH) and school term (ST). While visitor numbers in the kiwi house were not counted, visitor numbers at the zoo were over double during SH than in ST. Three days from each period were pseudorandomly selected for analysis, with each kiwi’s behaviours continuously sampled for 1hr 3 times a day. Behaviour between treatments was largely consistent except abnormal behaviour was higher in SH (33.12%) compared to ST (15.11%) and the kiwis spent more time resting in ST (22.82%) than in SH (5.04%). Time of day also influenced behaviour, with the kiwis showing higher levels of foraging at 10am (54.54%) compared to 1pm (46.40%) and 4pm (43.13%). Repetitive behaviour was highest at 1pm (3.72%) compared to 10am (0.11%) and 4pm (1.33%). Unfortunately, the video cameras did not capture all the kiwis’ behaviour, as they spent 42% of the total observation time out of sight. Our study suggests that visitors do have an impact on kiwi behaviour, which supports the finding of a previous study. Along with differences in behavioural patterns across time of day, these results provide information to care staff when considering management decisions. To our knowledge, this is the first study of captive kiwi behaviour across different times of day. More research is needed on potential visitor impacts across time of day, school period, and sex/age of birds
The impact of environmental information disclosure on the cost of green bonds in New Zealand
No full textSUB RESEARCH QUESTIONS
• Is there a relationship between well-structured disclosures and the cost of green bonds?
• Does the transparency in EID affect the cost of green bonds in the New Zealand bond market
Mapping the suitability range of Bactrocera dorsalis (Oriental Fruit Fly) with reference to New Zealand
No full textBactrocera dorsalis (Oriental Fruit Fly) is a widely known invasive species for its adaptability to different environments and ability to host over 250 fruits and vegetables. In January and February 2025, there were two instances where a male Oriental Fruit Fly was found and eradicated in the Auckland region, proving its potential and risk of establishing in New Zealand. This study aims to determine where the Oriental Fruit Fly is most likely to establish and spread in New Zealand using a species distribution model. By using ArcGIS Pro and Maxent, prediction models can be made to highlight the areas in New Zealand where they are likely to establish. Preliminary findings show that the Oriental Fruit Fly is most likely to establish within the North Island, specifically in the Auckland region, where the annual temperatures and precipitation are higher than in other parts of the country. It is vital to know where the Oriental Fruit Fly is most likely to spread and establish in New Zealand because if it establishes, there will be direct effects on New Zealand's biosecurity and our agricultural exports. Our study suggests that biosecurity measures and trap checks should be increased within the Auckland region, especially during the summer months, when the temperature is most optimal for the reproduction of the Oriental Fruit Fly
Hybrid deep learning approach for grape and apple leaf disease detection using CNN, YOLOv11 and EfficientNet-V2s
No full textRESEARCH QUESTIONS
1. How can the integration of CNN, YOLOv11, and EfficientNet-V2s improve the lo calisation and classification of grape and apple leaf diseases in complex agricultural environments?
2. Can the proposed hybrid model achieve real-time inference and high accuracy while remaining computationally efficient enough for deployment in resource-constrained settings?
3. How does the model perform when subjected to variable field conditions, such as inconsistent lighting, overlapping leaves, and background noise?
ABSTRACT
Grapevine and apple leaf diseases pose a substantial threat to global fruit production, affecting both crop productivity and quality. While early detection is crucial for effective management, traditional approaches such as manual inspection remain slow, subjective, and unsuitable for large scale deployment. This paper presents a hybrid deep learning architecture that combines classification and detection in a structured, modular pipeline. The proposed system integrates a CNN to distinguish between apple and grape leaves, YOLOv11 for rapid and precise localisation of diseased regions, and EfficientNet-V2s for fine-grained disease classification.
The framework was trained and tested on ten publicly available datasets comprising a diverse set of grape and apple leaf images under variable field-like conditions. Strong detection performance was achieved using YOLOv11, with a test mAP0.5 of 0.990 and high precision and recall, confirming its robustness across different visual environments. However, the final classification stage using EfficientNet-V2s yielded moderate results, with a test accuracy of 49.2%, highlighting challenges related to class imbalance and subtle inter-class similarities in visual symptoms.
Despite classification limitations, the modular structure of the system enabled effective disease region isolation and showed superior performance when compared to baseline models such as YOLOv8 + CNN and YOLOv5 + ResNet. The findings offer a practical and extensible foundation for automated crop disease monitoring. Future adaptations of this work can support improved classification accuracy, domain generalisation, and real world deployment in resource constrained agricultural environments