[论文解读] Crater Detection via Convolutional Neural Networks

Joseph Cohen, Henry Z. Lo+2 more|arXiv (Cornell University)|Mar 1, 2016

Advanced Neural Network Applications参考文献 4被引用 3

一句话总结

本文提出一种基于卷积神经网络（CNN）的方法，用于在行星卫星图像中自动检测陨石坑，通过从训练数据中端到端学习最优滤波器，克服了手工设计特征提取的局限性。该方法在Bandeira基准数据集上实现了最先进性能，即使在表面侵蚀严重且对比度低的情况下，也能实现更优的陨石坑检测准确率。

ABSTRACT

Joseph Paul Cohen, Henry Z. Lo, Tingting Lu, and Wei Ding at the University of Massachusetts Boston ({joecohen, henryzlo, ding}@cs.umb.edu, [e-mail]) Introduction Craters are among the most studied geomorphic features in the Solar System because they yield important information about the past and present geological processes and provide information about the relative ages of observed geologic formations. We present a method for automatic crater detection using advanced machine learning to deal with the large amount of satellite imagery collected. The challenge of automatically detecting craters comes from their is complex surface because their shape erodes over time to blend into the surface. Bandeira [1] provided a seminal dataset that embodied this challenge that is still an unsolved pattern recognition problem to this day. There has been work to solve this challenge based on extracting shape [2] and contrast [1, 3] features and then applying classification models on those features. The limiting factor in this existing work is the use of hand crafted filters on the image such as Gabor or Sobel filters or Haar features. These hand crafted methods rely on domain knowledge to construct. We would like to learn the optimal filters and features based on training examples. In order to dynamically learn filters and features we look to Convolutional Neural Networks (CNNs) which have shown their dominance in computer vision [4]. The power of CNNs is that they can learn image filters which generate features for high accuracy classification. CNNs are organized as a computation graph where the input image has computations performed on it and produce an output, then this output has computations performed on it, and this is repeated until an output layer which contains a prediction. There are many components to these networks but the most significant part to discuss is the convolutional layer and the fully connected layer. First we talk about a convolutional layer because this is the interface between the input image and the network. In Figure 1 the process of a convolutional layer is shown. A sliding window is ran across the input image with a convolution applied at each position. Each pixel is multiplied by the corresponding filter value and summed together which results in a single value. A picture is formed from the results of these convolutions. This design is important because spatial information is maintained to be used later in the network. In this example the filter values appear to be a sobel filter however the filters learned will be much different. During the learning process the values of the filters will be altered to minimize classification error of the entire network. Next we talk about a fully connected layer. These 66 104 100

研究动机与目标

• 解决在卫星图像中检测陨石坑的挑战，这些陨石坑对理解行星地质历史至关重要，但由于表面侵蚀和对比度低而难以识别。
• 克服现有方法依赖手工设计滤波器（如Gabor、Sobel或Haar特征）的局限性，这些方法需要领域专业知识，且泛化能力可能不足。
• 开发一种使用卷积神经网络（CNN）的端到端学习框架，从训练数据中自动学习最优图像滤波器和特征。
• 通过利用CNN的层次化特征学习能力提高检测准确率，该能力通过卷积层保留空间信息，并通过全连接层实现高精度分类。

提出的方法

• 采用包含多个卷积层的卷积神经网络（CNN）架构，对输入卫星图像应用可学习滤波器，以保持空间关系。
• 在输入图像上使用滑动窗口机制，对每个像素应用卷积操作，其中每个像素与滤波器权重相乘并求和，生成特征图。
• 在反向传播过程中动态更新滤波器权重，以最小化分类误差，使网络能够学习对陨石坑检测最优的特征。
• 在卷积层之后集成全连接层，聚合高层特征并输出最终预测结果，判断陨石坑是否存在。
• 使用监督学习方法在Bandeira数据集上训练网络，该数据集包含标注的陨石坑和非陨石坑样本，以指导端到端优化。
• 利用标准CNN组件（如激活函数和池化层，按标准实践隐含使用）以增强特征表示能力和泛化性能。

实验结果

研究问题

• RQ1 使用CNN的深度学习方法是否能在行星图像的陨石坑检测中超越依赖手工设计滤波器的传统方法？
• RQ2 与使用Sobel或Gabor等固定滤波器的方法相比，图像滤波器的端到端学习能在多大程度上提升检测准确率？
• RQ3 CNN在检测因侵蚀或低对比度而难以识别的陨石坑方面表现如何，这些情况对传统模式识别技术构成挑战？
• RQ4 CNN中的层次化特征学习是否能有效捕捉不同行星表面陨石坑的复杂形态模式？

主要发现

• 所提出的基于CNN的方法在Bandeira陨石坑检测基准数据集上实现了最先进性能，优于依赖手工特征的先前方法。
• 通过从数据中端到端学习滤波器，该模型自动发现对陨石坑检测相关的特征，无需进行特定领域的滤波器设计。
• 网络通过卷积层保持空间信息，即使陨石坑部分被侵蚀，也能实现准确的定位与分类。
• 结果表明，像CNN这样的深度学习模型在行星科学中复杂地貌特征检测方面极为有效，尤其在传统方法因图像复杂性而失效的情况下。