LeAF: Leveraging Deep Learning for Agricultural Pest Detection and Classification for Farmers

Farmers face many challenges while growing crops such as monitoring and maintaining plant health. Key indicators of poor plant health are plant anomalies such as pests, plant disease, and weeds, which can decrease crop yield. Over 40% of global crop production is lost to plant anomalies, costing $220 billion annually. As the global population and demand for food increases, farmers will have to grow more food, making manual surveying for plant anomalies increasingly difficult. This forces farmers to excessively and indiscriminately apply more fertilizers and pesticides across their whole fields, often to both healthy and unhealthy plants, unnecessarily wasting acres worth of chemicals and increasing chemical contamination of food and environmental footprint of agriculture as the chemicals release greenhouse gases after their application and leak into ecosystems. Recent advances in deep learning with Convolutional Neural Networks (CNNs) allow using imaging data to solve this problem. LeAF aims to provide farmers with an end-to-end system to survey crops on the field and take targeted actions to maintain plant health. By focusing on agricultural pests, this paper demonstrates the following capabilities for the visual perception sub-system of LeAF: (1) use CNNs on field images to get plant-specific data with bounding box based detection and classification about plant anomalies at human-level accuracy and (2) combine detection and classification functionality into a single compact distilled model that can run on farmer accessible mobile phones or in embedded devices in agricultural tractors and robots with low latency and high throughput to enable real-time processing on video feeds. With lightweight and accurate plant anomaly detection and classification, LeAF addresses plant health management challenges faced by farmers, empowering them with actionable insights to enhance productivity while minimizing chemical usage and its environmental impact.

---

Thematic Sessions: Short Papers