Bird Detection in Agriculture Environment using Image Processing and Neural Network

This paper addresses the problem of detecting pest birds in the agriculture field. So far, various research methods have been proposed for bird detection. However, bird detection in a low-lying field is still challenging since it is difficult to distinguish objects in the dynamic background including moving leaves and clouds. To recognize objects in the dynamic background, deep learning has been used due to its efficiency. However, detecting objects in a video with deep learning is not fast enough since the process needs high computational cost. In addition, if the object is tiny, the accuracy becomes lower. To solve these problems, we propose adapting deep learning to certain cropped small areas of a frame where there is a high possibility of bird presence, based on the result of the image processing. First, the moving objects are extracted by conducting the background subtraction based on Gaussian Mixture Model. Then, color extraction and median filter remove the unnecessary objects in the agricultural environment. Lastly, the minimized moving objects are classified with the neural network object classifier. The experimental results demonstrate that applying neural networks to specific areas of an image brings a higher accuracy than the whole original frame.