改进型卷积神经网络无人机工地识别

Construction site recognition of UAV based on improved convolutional neural network

现有通用目标检测算法在无人机工地识别任务中容易产生精度低下等问题,针对该问题,该文提出一种卷积神经网络模型,用于复杂环境下相似目标检测。该模型首先利用无人机高空拍摄图片作为数据集,通过高斯模糊、图像变换等方法进行数据增强,为模型泛化能力的提高提供数据支撑。然后基于Darknet-53特征提取网络实现多尺度特征融合,通过在网络模型中添加SPP-net(spatial pyramid pooling networks)应对模型中特征易消失问题。最后优化损失函数,解决模型正负样本不均衡问题。实验结果证明该模型mAP值达到84.94%,可为城市内土地规划、施工和违章搭建监管等领域提供技术支撑。

Existing general target detection algorithms are prone to produce low accuracy problems in UAV site recognition tasks.To solve this problem,this paper proposes a convolutional neural network model for similar target detection in complex environments.Firstly,aerial images taken by UAV are used as the data set,and the data are enhanced by Gaussian blur and image transformation,which provide data support for improving the generalization ability of the model.Then,multi-scale feature fusion is realized based on Darknet-53 feature extraction network,and SPP-net(spatial pyramid pooling networks)is added to the network model to solve the problem of feature disappearing easily.Finally,the loss function is optimized to solve the imbalance of positive and negative samples.Experimental results show that the mAP value of the model reaches 84.94%,which can provide technical support for urban land planning,construction and supervision of illegal construction.