Combined with DCT, SPIHT and ResNet
to Identify Ancient Building Cracks
from Aerial Remote Sensing Images
Tiantian Tang, Bingcai Chen
(&) , and Ruiheng Hu
Xinjiang Normal University, Urumqi, China
528879201@qq.com,33682049@qq.com
Abstract. In order to study the crack detection of ancient buildings in aerial
remote sensing images, this paper proposes to preprocess remote sensing images
by combining block DCT transform with SPIHT algorithm terminating at the
threshold to retain the target information to the maximum extent and reduce the
number of processed images. Then, ResNet model is used to detect the cracks in
ancient buildings.
Keywords: SPIHT Á DCT Á ResNet Á Crack detection Á Aerial remote sensing
images
1 Introduction
The traditional manual inspection of ancient buildings has obviously failed to meet the
needs of comprehensive inspection and restoration of ancient buildings. The characteristic of remote sensing technology is that it does not need to detect the target on the
spot, but can directly collect and identify relevant information. UAV remote sensing
image has the characteristics of wide coverage, large change of gray value, complex
and changeable texture information, and precious acquired information. Therefore, how
to deal with massive remote sensing images, reduce training examples and maximize
the retention of feature information, so as to facilitate classification is the research
focus.
Discrete cosine transform (DCT), first proposed by Ahmed and Rao, is considered
as the best method to transform language signal and image signal because its performance is close to that of k-l transform. DCT transform can concentrate image energy to
a few low-frequency DCT coefficients, which has low complexity and strong generalization ability [1]. DCT uses Huffman coding, and the frequency determines the
length of coding. The transformed image is not dynamically encoded and the image
compression ratio cannot be customized according to the actual research object. Set
Partitioning in Hierarchical Trees (SPIHT) algorithm determines the initial scanning
threshold based on the image wavelet coefficient [2]. The initial threshold value is
changed dynamically according to the last scan result, and the threshold value is halved
on the basis of the last scan. After setting the compression ratio, if the scanning
threshold is reduced to the expected value, the cyclic scanning will stop. Thus, the
© Springer Nature Singapore Pte Ltd. 2020
Q. Liang et al. (Eds.): Artificial Intelligence in China, LNEE 572, pp. 313–318, 2020.
https://doi.org/10.1007/978-981-15-0187-6_36
to Identify Ancient Building Cracks
from Aerial Remote Sensing Images
Tiantian Tang, Bingcai Chen
(&) , and Ruiheng Hu
Xinjiang Normal University, Urumqi, China
528879201@qq.com,33682049@qq.com
Abstract. In order to study the crack detection of ancient buildings in aerial
remote sensing images, this paper proposes to preprocess remote sensing images
by combining block DCT transform with SPIHT algorithm terminating at the
threshold to retain the target information to the maximum extent and reduce the
number of processed images. Then, ResNet model is used to detect the cracks in
ancient buildings.
Keywords: SPIHT Á DCT Á ResNet Á Crack detection Á Aerial remote sensing
images
1 Introduction
The traditional manual inspection of ancient buildings has obviously failed to meet the
needs of comprehensive inspection and restoration of ancient buildings. The characteristic of remote sensing technology is that it does not need to detect the target on the
spot, but can directly collect and identify relevant information. UAV remote sensing
image has the characteristics of wide coverage, large change of gray value, complex
and changeable texture information, and precious acquired information. Therefore, how
to deal with massive remote sensing images, reduce training examples and maximize
the retention of feature information, so as to facilitate classification is the research
focus.
Discrete cosine transform (DCT), first proposed by Ahmed and Rao, is considered
as the best method to transform language signal and image signal because its performance is close to that of k-l transform. DCT transform can concentrate image energy to
a few low-frequency DCT coefficients, which has low complexity and strong generalization ability [1]. DCT uses Huffman coding, and the frequency determines the
length of coding. The transformed image is not dynamically encoded and the image
compression ratio cannot be customized according to the actual research object. Set
Partitioning in Hierarchical Trees (SPIHT) algorithm determines the initial scanning
threshold based on the image wavelet coefficient [2]. The initial threshold value is
changed dynamically according to the last scan result, and the threshold value is halved
on the basis of the last scan. After setting the compression ratio, if the scanning
threshold is reduced to the expected value, the cyclic scanning will stop. Thus, the
© Springer Nature Singapore Pte Ltd. 2020
Q. Liang et al. (Eds.): Artificial Intelligence in China, LNEE 572, pp. 313–318, 2020.
https://doi.org/10.1007/978-981-15-0187-6_36