Crater recognition on Mars is of paramount importance for many space science applications, such as accurate planetary surface age dating and geological mapping. Such recognition is achieved by means of various image-processing techniques employing traditional CNNs (convolutional neural networks), which typically suffer from slow convergence and relatively low accuracy. In this paper, we propose a novel CNN, referred to as MC-UNet (Martian Crater U-Net), wherein classical U-Net is employed as the backbone for accurate identification of Martian craters at semantic and instance levels from thermal-emission-imaging-system (THEMIS) daytime infrared images. Compared with classical U-Net, the depth of the layers of MC-UNet is expanded to six, while the maximum number of channels is decreased to one-fourth, thereby making the proposed CNN-based architecture computationally efficient while maintaining a high recognition rate of impact craters on Mars. For enhancing the operation of MC-UNet, we adopt average pooling and embed channel attention into the skip-connection process between the encoder and decoder layers at the same network depth so that large-sized Martian craters can be more accurately recognized. The proposed MC-UNet is adequately trained using 2∼32 km radii Martian craters from THEMIS daytime infrared annotated images. For the predicted Martian crater rim pixels, template matching is subsequently used to recognize Martian craters at the instance level. The experimental results indicate that MC-UNet has the potential to recognize Martian craters with a maximum radius of 31.28 km (136 pixels) with a recall of 0.7916 and -score of 0.8355. The promising performance shows that the proposed MC-UNet is on par with or even better than other classical CNN architectures, such as U-Net and Crater U-Net.
22 December 2023 in International Journal of Applied Earth Observation and Geoinformation
This paper proposes an encoding–decoding network with a pyramidal representation module, which will be referred to as EDPNet, and is designed for efficient semantic image segmentation. On the one hand, during the encoding process of the proposed EDPNet, the enhancement of the Xception network, i.e., Xception+ is employed as a backbone to learn the discriminative feature maps. The obtained discriminative features are then fed into the pyramidal representation module, from which the context-augmented features are learned and optimized by leveraging a multi-level feature representation and aggregation process. On the other hand, during the image restoration decoding process, the encoded semantic-rich features are progressively recovered with the assistance of a simplified skip connection mechanism, which performs channel concatenation between high-level encoded features with rich semantic information and low-level features with spatial detail information. The proposed hybrid representation employing the proposed encoding–decoding and pyramidal structures has a global-aware perception and captures fine-grained contours of various geographical objects very well with high computational efficiency. The performance of the proposed EDPNet has been compared against PSPNet, DeepLabv3, and U-Net, employing four benchmark datasets, namely eTRIMS, Cityscapes, PASCAL VOC2012, and CamVid. EDPNet acquired the highest accuracy of 83.6% and 73.8% mIoUs on eTRIMS and PASCAL VOC2012 datasets, while its accuracy on the other two datasets was comparable to that of PSPNet, DeepLabv3, and U-Net models. EDPNet achieved the highest efficiency among the compared models on all datasets.