基于U-Net网络的鸟群图像分割研究

Research on bird Image segmentation based on U-Net

当前在通用航空活动中对于鸟群活动情况的了解还是依靠飞行员对环境的观察,并且国内外驱鸟方式并不适合大范围推广使用,着眼于以上问题,利用基于深度学习的图像分割方法来解决该问题,将U-Net图像分割策略引入航空图像分析中,通过深度卷积神经网络对鸟群活动图像进行处理;针对图像背景复杂、目标分布不均匀且大小不定等特点,提出优化U-Net网络,以更好地保存和利用图像信息。实验证明该优化网络图像分割效果优于现有方法。

Currently,the spotting of birds activity mainly relys on the pilots' observation in the general aviation activities.Besides,the methods at home and abroad are not suitable for widespread use.Therefore,focusing on the problem above,with the aid of image segmentation method based on the deep learning,the problem can be solved.U-Net image segmentation strategy is introduced to aviation images analysis,and deep convolutional neural network is used to process the images of birds.Based on the characteristics of complex image background,uneven target distribution and variable size,an optimized U-NET network is proposed to save and utilize the image information better.Experiments show that the image segmentation effect of the optimized network is better than existing method.