Development of Machine Learning Model for Crop Pests and Diseases Diagnosis Based on Crop Imagery Data
- Name of Principal Investigator: Dr. Hudson Laizer
- Grant Amount: US$ 51,000.00 (US Dollar Fifty-one thousand only)
- Grant Award Number: AFS-2504001568
- Country of Implementation: Tanzania
- Interest: Sustainable agriculture, Agro-ecology, Climate smart agriculture, Biodiversity conservation, Botanical pesticides, Pre- and Post-harvest loss in agriculture
Goals
The goal of the project is to develop a Machine Learning models for early detection of Common bean and Irish potato diseases in the Southern Highlands regions of Tanzania.
Objectives
The specific objectives of the study are to:
- To assess farmers' knowledge and perceptions on Artificial Intelligence, Machine Learning, pests and disease management in Common bean and Irish potato farming.
- To collect high-quality imagery data for healthy and diseased Common bean and Irish potato crops.
- To annotate and prepare the collected image data for Machine Learning models development.
- Train and build capacities of farmers, agricultural extension officers and other stakeholder.
Key activities
- The key activities for the six months of the project where: Inception meeting for project stakeholders to communicate the project objectives, responsible AI practices, risks and
- Enrolment of farmers to the project with the goal of Identifying with the community partners
- Development of data collection tool using ODK
- Training of project team and enumerators on using digital data collection tools
- Initial data collection for tool testing and verification
Key Achievements
- The project team has collected field data and are verifying the uploaded data. The target was 120,000 but have managed to collect 129,500 images for all classes. The dataset that has been pre-processed for machine learning is a total of 36,000 images. The data exploration activity involved baseline training of models for image classification and object detections tasks. 500 images were used from each class to ensure all classes are balanced. The standard GPU on Google colab environment was used to train the baseline models.
- The project PI (Dr. Hudson Laizer) and ML expert (Dr. Neema Mduma) presented the project work in Black in AI workshop co-located with NeurIPS 2022 Conference in New Orleans, Louisiana, United States of America. The workshop enabled the project team to establish both formal and informal networks with other researchers in the field of AI and ML for future collaborations.
Next Steps
The image pre-processing work is ongoing to reach the target of 120,000+ images for six classes. The dataset that will be shared in ATPS open repositories. The project team will train more model architectures for image classification and object detection for suitable model selection on deployment on mobile application. After model development and testing, the project team will organize workshops, training and dialogues where participants including smallholder farmers, government officials, private sectors etc. will be engaged to discuss matters of mutual benefits. The engagement of multiple stakeholders during the dissemination of project results is important not only in sharing project outcomes but also in collecting the diverse opinions that will help to identify any need for policy review and change 2.2.4 Gender equality and inclusion action