Introduction
-
• Based on Convolutional Neural Network and Dense Block, we propose a lightweight densely connected network-AD-Net for the extraction of open-pit coal mining areas from Sentinel-2 re-mote sensing images.
-
• The AD-Net model consists of two convolutional layers, two pooling layers, a channel attention module and a Dense Block. The two convolution-al layers greatly reduce the complexity of the model, and the Dense Block enhances the feature propagation while reducing the parameter computation.
-
• A deep neural framework is designed for edge cloud with different modules that runs independently and communicate with each other to improve the processing efficiency.
-
• Furthermore, we create and build a unique remote sensing image service system that connects many datacentres and their associated edge networks, employing the edge-cloud architecture.
Related work
Materials and methods
Materials
Study area
Dataset description
Dataset processing
Methods and techniques
The CNN Model
Dense block and its architecture
Channel attention module
The proposed AD-Net model
Loss function and evaluation metrics
Precision
Recall
F1-Score
OA (Overall Accuracy)
Kappa (Kappa Coefficient)
Edge-cloud for distributed CNN model
Results and discussion
-
Part I: The optimal slice size of the input data and the optimal optimizer of the model are determined.
-
Part II: The optimal model is compared with the advanced CNN and RNN models, and Recall, Precision, F1, OA, and kappa are used as evaluation metrics.
Slice experiments
Slice size | Recall | Precision | F1-Score | OA | Kappa |
---|---|---|---|---|---|
9 × 9 | 0.975 | 0.897 | 0.934 | 0.931 | 0.861 |
17 × 17 | 0.934 | 0.958 | 0.946 | 0.947 | 0.908 |
33 × 33 | 0.974 | 0.914 | 0.943 | 0.941 | 0.881 |
Slice size | Recall | Precision | F1-Score | OA | Kappa |
---|---|---|---|---|---|
9 × 9 | 0.959 | 0.927 | 0.942 | 0.941 | 0.882 |
17 × 17 | 0.967 | 0.937 | 0.952 | 0.951 | 0.902 |
33 × 33 | 0.931 | 0.945 | 0.938 | 0.934 | 0.889 |
Slice size | Recall | Precision | F1-Score | OA | Kappa |
---|---|---|---|---|---|
9 × 9 | 0.937 | 0.947 | 0.942 | 0.942 | 0.885 |
17 × 17 | 0.943 | 0.961 | 0.952 | 0.952 | 0.905 |
33 × 33 | 0.886 | 0.964 | 0.923 | 0.927 | 0.898 |
Experimental parameters
Slice size | Recall | Precision | F1-Score | OA | Kappa |
---|---|---|---|---|---|
SGD | 0.949 | 0.953 | 0.951 | 0.951 | 0.902 |
AdaGrad | 0.968 | 0.907 | 0.936 | 0.934 | 0.867 |
RMSprop | 0.904 | 0.963 | 0.933 | 0.935 | 0.871 |
Adma | 0.970 | 0.945 | 0.957 | 0.957 | 0.913 |
Slice size | Recall | Precision | F1-Score | OA | Kappa |
---|---|---|---|---|---|
SGD | 0.970 | 0.948 | 0.959 | 0.958 | 0.916 |
AdaGrad | 0.958 | 0.936 | 0.947 | 0.946 | 0.892 |
RMSprop | 0.944 | 0.956 | 0.950 | 0.950 | 0.900 |
Adma | 0.986 | 0.938 | 0.961 | 0.960 | 0.920 |
Slice size | Recall | Precision | F1-Score | OA | Kappa |
---|---|---|---|---|---|
SGD | 0.987 | 0.935 | 0.960 | 0.959 | 0.918 |
AdaGrad | 0.927 | 0.953 | 0.940 | 0.941 | 0.882 |
RMSprop | 0.974 | 0.938 | 0.955 | 0.955 | 0.909 |
Adma | 0.953 | 0.964 | 0.959 | 0.959 | 0.918 |
Comparison of different CNN models
Net | Recall | Precision | F1-Score | OA | Kappa |
---|---|---|---|---|---|
VGG-16 | 0.470 | 0.991 | 0.638 | 0.733 | 0.503 |
AlexNet | 0.997 | 0.817 | 0.898 | 0.887 | 0.766 |
GoogLeNet | 0.973 | 0.904 | 0.034 | 0.912 | 0.824 |
ResNet50 | 0.927 | 0.959 | 0.942 | 0.943 | 0.887 |
Xception | 0.983 | 0.913 | 0.947 | 0.945 | 0.889 |
DenseNet121 | 0.975 | 0.926 | 0.950 | 0.949 | 0.896 |
AD-Net | 0.953 | 0.964 | 0.959 | 0.959 | 0.918 |
Results of AD-Net model over the edge-cloud setup
CPU model | Speed (GHz) | Cores | ECUs | Memory (GB) | Storage (TB) |
---|---|---|---|---|---|
Servers | |||||
E5645 | 2.400 | 12 | 18.8 | 32 | 8 |
Virtual Machines | |||||
t1.tiny | 800 MIPS | 1 | 1 | 0.6 | 650 MB |
t2.micro | 1200 MIPS | 1 | 1 | 0. 613 | 1 GB |
CLOUD | EDGE-CLOUD | |||
---|---|---|---|---|
Training time (s) | Prediction time (s) | Training time (s) | Prediction time (s) | |
CNN | 10,543.6 | 4598.34 | 8451.6 | 1297.12 |
RNN | 10,751.5 | 4384.4 | 8341.66 | 1188.3 |
U-net | 10,734.95 | 4381.64 | 8344.01 | 1131.53 |