Using Machine Learning Techniques In The Visual Detection Of Starlings In Vineyards

Abstract

This paper deals with the visual detection of starlings. The aim is to design an earlywarning detection system that protects crops from flocks of starlings. This system uses computervision and machine learning algorithms. In the first phase, the activity in the vineyard was collected.Further, the neural network model using a cloud-based AutoML platform was trained and evaluated.The final classifier distinguishes objects into several categories. These categories include individualbirds, flocks, and various unintended objects such as flies and bees. Overall, the flock detectionalgorithm achieved 89 % accuracy and 94% recall.