Abstract Urban images usually contain buildings, pedestrians, vehicles, roads, and other complex objects. Due to this high complexity, object detection and classification of urban images is a challenging task. In this paper, a novel convolutional neural network named complex background classification network (CBCNet) is proposed for urban image classification. Unlike the existing image classification and object detection methods such as residual network (ResNet) and faster region-based convolutional neural networks (Faster R-CNN), CBCNet first utilizes a multilayer perceptron convolutional layer instead of a linear convolutional layer to extract representative features from urban images, and then uses a backpropagation neural network to optimize the extracted object parameters. In addition, we build a standard urban image dataset (UID) which contains eight categories. Qualitative experiments on two benchmark datasets demonstrate that classification accuracy and computation of CBCNet outperform the state-of-the-art methods.

中文翻译：

---

复杂背景分类网络：一种城市图像分类的深度学习方法

摘要 城市图像通常包含建筑物、行人、车辆、道路等复杂物体。由于这种高复杂性，城市图像的目标检测和分类是一项具有挑战性的任务。在本文中，提出了一种名为复杂背景分类网络（CBCNet）的新型卷积神经网络用于城市图像分类。与现有的图像分类和目标检测方法如残差网络 (ResNet) 和更快的基于区域的卷积神经网络 (Faster R-CNN) 不同，CBCNet 首先利用多层感知器卷积层而不是线性卷积层来提取具有代表性的特征。城市图像，然后使用反向传播神经网络优化提取的对象参数。此外，我们构建了一个标准的城市图像数据集（UID），其中包含八个类别。在两个基准数据集上的定性实验表明，CBCNet 的分类精度和计算性能优于最先进的方法。