Comparison of estimating vegetation index for outdoor free-range pig production using convolutional neural networks
Abstract
The objective of this study was to predict the change in occupancy rate of corn caused by the grazing of twenty gestating sows in a mature cornfield by capturing images with a camera-equipped unmanned aerial vehicle (UAV). A large amount of data is required to effectively train convolutional neural network (CNN)-based deep learning, However, the UAV only captured a limited number of images so a data augmentation method that could effectively increase the data was proposed. Various CNNs were used as regression networks for comparison, and the applicability and scalability of deep learning were verified. Ten corn field images were captured by a UAV over approximately two weeks during which gestating sows were allowed to graze freely on a corn field measuring 100×50 m<sup>2</sup>. The images were corrected to a bird's-eye view, and then divided into 32 segments and sequentially inputted into the different CNN detectors to detect the corn images according to their condition. A total of 43 raw training images selected randomly from the 320 segmented images were flipped to create 86 images, and then these images were further augmented by rotating them in 5-degree increments to create a total of 6,192 images. These 6192 images were further augmented by applying three random color transformations to each image, resulting in a dataset of 24,768 images. Of the tested networks, AlexNet, Vgg16, and Vgg19 showed higher prediction accuracy than the other networks. GoogLeNet showed slightly higher occupancy rates of damaged corn (ACD) on day 3 and day 5 than on day 4, and slightly higher rates of corn with stubble (ACS) and corn in all conditions (ACT) on day 11 and day 12 compared to day 10. ResNet50 and ResNet101 generally showed good prediction accuracy, but their predictions were slightly higher on day 14. Overall, the CNNs demonstrated excellent prediction accuracy, confirming the potential and scalability of deep learning. The proposed method has effectively estimated the occupancy rate using a limited number of cornfield photos, and there is a high potential for expanding it into other areas of livestock farming in the future.