David Alvear

CropCount project

The CropCount project, initiated in collaboration with a distinguished company, aimed to automate the labor-intensive task of counting elements in drone-captured images. Addressing both the inefficiencies and health implications of manual counting, this innovation significantly reduced processing time and mitigated repetitive stress injuries. The software not only achieved counting speeds surpassing human capabilities but also generated CSV files detailing the coordinates of detected elements. Initially employing traditional computer vision tools like OpenCV, Skimage, and numpy for image processing and element recognition, a subsequent version integrated deep learning with TensorFlow and Keras to enhance accuracy. This upgraded design utilized a dataset of approximately 100,000 drone-captured crop images and incorporated GPU-based processing for swift analysis. The final workflow, supported by an Nvidia GPU GTX-2080, featured GUI tools, sliding window image evaluations, neural network classification, and sophisticated filtering methods to pinpoint plant locations with over 90% accuracy, even amid varying lighting and cloud conditions. This transformative solution, showcased below, not only fulfilled its efficiency objectives but also exemplifies innovative research aimed at addressing real-world challenges.