Computer vision for plant growth management
Loading...
Date
2024
Journal Title
Journal ISSN
Volume Title
Publisher
UMT, Lahore
Abstract
The idea for our project emerged after witnessing a relative struggling with crop health management last season. In Pakistan, many farmers rely on traditional, manual methods to assess plant health—typically walking through fields and visually inspecting plants. This approach is not only labor-intensive but also prone to missing early signs of disease. To address this, we developed an affordable, IoT-based computer vision system using a Raspberry Pi paired with a basic camera module. The total cost of the setup was approximately PKR 40,000, making it significantly more accessible compared to the high-end equipment used on large-scale farms. The system captures images of crops and analyzes them to detect signs of disease, such as leaf spots or discoloration. Initially, we designed a complex dashboard with graphs and percentages, but field testing revealed that it was too difficult to use for those unfamiliar with digital tools. We redesigned it with a much simpler interface that provides a basic status—"Plant Healthy" or "Plant Sick"—along with brief suggestions for action. The system was tested on a local farm for three weeks in February. Results were mixed: it successfully detected diseases like potato blight early in some cases, but struggled with environmental challenges such as poor lighting and dust accumulation on the lens, which we hadn't initially accounted for. Currently, the detection accuracy is around 75–80%. While not perfect, it still represents an improvement over traditional methods. Work is ongoing to improve the model and address the environmental limitations, with plans for continued development in future..