桃色视频

Pollinators such as bees, butterflies and hoverflies are essential for ecosystem functioning and food security, contributing to the pollination of over 1/3 of global food crops (Klein et al., 2006; Ollerton et al., 2011). However pollinator populations are in significant decline due to a combination of anthropogenic pressures including habitat loss, pesticide use, climate change and disease (Potts et al., 2016; Rundlöf et al., 2015). This decline poses a substantial threat to ecosystem resilience and agricultural productivity. Despite their ecological and economic importance, current pollinator monitoring methods remain labour-intensive, spatially constrained and temporally limited (O’Connor et al., 2019). There is a pressing need for scalable, cost-effective and accurate tools to monitor pollinator populations and inform conservation and land management strategies.  

This project aims to develop and deploy an AI-driven image analysis tool to automate the identification and quantification of insect pollinators from large-scale photographic and video datasets. The research will integrate ecological fieldwork, computer vision and stakeholder engagement to: 

1.Develop and optimise deep learning models for pollinator identification and abundance estimation using annotated image datasets. 

2.Validate model performance across diverse habitats, including agricultural and semi-natural landscapes. 

3.Develop ML models to predict the occurrence or frequency of interactions between pollinators and plant species.  

4.Co-design a user-friendly software interface with input from researchers, conservation practitioners, farmers and citizen scientists.  

The project will leverage Cranfield University’s Living Laboratory and Urban Observatory for image collection and stakeholder engagement. The resulting tool will support national biodiversity monitoring efforts and contribute to policy development through evidence-based insights into pollinator dynamics. By enabling high-throughput, standardised and accessible pollinator monitoring, our platform has the potential to transform how we assess pollinator health, evaluate the impacts of environmental change and design effective conservation interventions.  


Methodology:

The project will adopt a mixed-methods approach: 

- Data collection: Deploy camera traps and artificial flower attractants across urban and agricultural sites to capture pollinator activity. 

- Model development for Pollinator Monitoring: Train and optimise deep learning models for pollinator detection and classification using annotated image datasets. Post-processing object tracking algorithms will be incorporated on the acquired videos to provide abundance estimation. 

- Model Development for Plant-Pollinator Interactions Prediction: The information-rich datasets acquired during the project will also be used to develop ML models such as Random Forest and Neural Networks to help understand and predict pairwise interactions between pollinators and plant species.  

- Software Engineering: integrate models into a standalone application with automated image processing, statistical analysis and reporting features. 

- Stakeholder Engagement: conduct workshops and interviews with conversation organisations, farmers and researchers to co-design the tool and ensure usability. 

Validation: compare model outputs with manual field surveys to assess accuracy and reliability. 

Partners and collaboration:

The project will engage with the UK Pollinator Monitoring Scheme (PoMS) and DEFRA’s Pollinator Advisory Group to ensure alignment with national biodiversity goals. 

Possible timeline: 

Year 1:  

• Literature review and stakeholder mapping. 

• Image collection setup and data acquisition. 

• Training in ecological field identification. 

• Initial model training and software prototype development. 

Year 2: 

• Expanded image collection across multiple sites. 

• Model refinement and validation. 

• Stakeholder workshops and interface co-design. 

Year 3:  

• Final software development and testing. 

• Dissemination via policy briefs, academic publications and public engagement. 

• Thesis writing and submission.