Crop Care (DeepConc-Lens)

Project Summary

Plant diseases and pests are a menace to farming systems in Africa, particularly in the subsaharan Africa. This is coupled by several factors including the conducive weather patterns which favor their propagation. Unfortunately, disease pressure does not match survailance tool availability, as well as limited extension workers. In this project, we aimed at empowering the farmers to diagnose plant diseases and pests as well as seek assistance from experts on available strategies to manage the identified pest and disease problem. We harness data and mordern computation using machine learning and computer intelligence to train crop-specific challenges and develop models for accurate diagnosis of the plant/crop challenges. Farmers are our main target and by giving them this tool, we are filling the existing gap in extension service provision which is the main challenge. We anticipate a challenge of limited internet connectivity, as well as limited availability of smartphones which are central to the consumption of this tool.

Focus

The primary objective of this project was to develop a web-based tool that could analyze uploaded plant images and accurately detect diseases and potential pest symptoms. The tool's seamless integration into agricultural practices facilitates the early detection of diseases, leading to reduced crop loss and enhanced food security.

How are pest and disease challenges identified

The tool employs a state-of-the-art convolutional neural network (CNN) model (RESNET50), trained on a meticulously curated dataset of diseased and healthy plant leaves. The model was fine-tuned through multiple training iterations to achieve high accuracy in disease identification. The tool extracts vital features from leaf images using image processing techniques  and employs these features to make informed disease predictions.

Training summarized report URL - https://api.wandb.ai/links/jumashafara/kry8zoqp

Features and Functionality:

Users can effortlessly upload images of plant leaves through an intuitive user interface. The tool processes these images, highlighting regions of interest and indicating the presence of diseases. Upon analysis completion, users receive predictions and recommendations for potential treatments, enabling informed decision-making.

Web tool URL - Deep Conc-Lens

Results and Impact:

During testing and validation, the model achieved an impressive accuracy rate of over 93% in disease detection across a diverse range of plant species including Tomato, Corn and Pepper. This accuracy underscores the tool's potential to revolutionize agricultural practices, minimizing the economic and ecological consequences of unchecked plant diseases. By facilitating early disease intervention, the tool holds the promise of significantly improving crop yields and the sustainability of farming practices.

Future Enhancements

Looking ahead, the project envisions expanding the tool's capabilities to detect a wider array of plant diseases. Additionally, user feedback will be instrumental in refining the user experience and incorporating new features. Integration with mobile devices and real-time disease monitoring systems is also under consideration to extend the tool's accessibility and usability.

Conclusion:

The Disease Detection Web Tool for Plant Leaves represents an advancement in the field of agricultural technology. By combining sophisticated machine learning techniques with user-centric design, the tool equips stakeholders with a powerful means to combat plant diseases effectively. As we continue to enhance and refine this tool, its impact on global food security and sustainable agriculture is poised to be substantial.