A Comparison of Image Processing Techniques for Bird Recognition

详细信息    查看全文

• 作者：Uma D. Nadimpalli ; Randy R. Price ; Steven G. Hall ; and Pallavi Bomma
• 刊名：Biotechnology Progress
• 出版年：2006
• 出版时间：February 2006
• 年：2006
• 卷：22
• 期：1
• 页码：9 - 13
• 全文大小：209K
• 年卷期：v.22,no.1(February 2006)
• ISSN：1520-6033

文摘

Bird predation is one of the major concerns for fish culture in open ponds. A novel method fordispersing birds is the use of autonomous vehicles. Image recognition software can improvetheir efficiency. Several image processing techniques for recognition of birds have been tested.A series of morphological operations were implemented. We divided images into 3 types, Type1, Type 2, and Type 3, based on the level of difficulty of recognizing birds. Type 1 imageswere clear; Type 2 images were medium clear, and Type 3 images were unclear. Localthresholding has been implemented using HSV (Hue, Saturation, and Value), GRAY, and RGB(Red, Green, and Blue) color models on all three sections of images and results were tabulated.Template matching using normal correlation and artificial neural networks (ANN) are the othermethods that have been developed in this study in addition to image morphology. Templatematching produced satisfactory results irrespective of the difficulty level of images, but artificialneural networks produced accuracies of 100, 60, and 50% on Type 1, Type 2, and Type 3images, respectively. Correct classification rate can be increased by further training. Futureresearch will focus on testing the recognition algorithms in natural or aquacultural settings onautonomous boats. Applications of such techniques to industrial, agricultural, or related areasare additional future possibilities.