Search for a command to run...
Drones and artificial intelligence (AI) offer a transformative approach to crop disease detection, enabling rapid and precise analysis beyond traditional methods. In this study, we developed and programmed a custom-built drone equipped with four 1500 kV brushless motors, a flight controller, 60-amp ESC, GPS, and a DJI O3 video transmission system. The hardware was assembled onto a carbon fiber frame with 3D-printed Thermoplastic Polyurethane (TPU) components, and a Raspberry Pi 5 was integrated for real-time inferencing. The AI model for crop disease detection was trained using drone collected data and publicly available datasets from Kaggle, leveraging PyTorch with Python3 on an HPC Cluster with an Nvidia H100 GPU and AMD EPYC 7453 CPU. Autonomous drone flight paths were planned using Litchi waypoint navigation software to create an autonomous point-to-point navigation flight path. Using the training data a Convolutional Neural Network (CNN) vanilla model was developed for training and inferencing the crop diseases from the images used in the dataset. Four major crops-corn, potato, rice, and wheat- covering 10 distinct diseases were analyzed. The model achieved an accuracy between 75% and 95% for most cases, except for rice crops, where performance was lower due to the thin leaves providing fewer distinguishing features. This issue can be mitigated with additional image augmentation and fine-tuning the CNN model. Although several challenges were encountered in securing access to local farms for large-scale testing, preliminary evaluations in controlled environments-including select accessible field and park settings-yielded promising results. Field testing confirmed the drone's ability to capture high-quality images even in light rain and cloudy conditions. The system, which costs under $ 500, achieved a maximum flight time of 27 minutes, offering a cost-effective alternative to traditional soil sensors and soil culture testing. The early identification of crop diseases enables targeted interventions, conserving resources, reducing excessive pesticide use, and increasing crop yields.