Figure 1.
Instances of different types of rocks and samples in the laboratory. For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr).
Figure 1.
Instances of different types of rocks and samples in the laboratory. For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr).
Figure 2.
Illustration of spectral curves corresponding to rocks and minerals. For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr).
Figure 2.
Illustration of spectral curves corresponding to rocks and minerals. For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr).
Figure 3.
A pipeline showing the developed rock and mineral classification approach.
Figure 3.
A pipeline showing the developed rock and mineral classification approach.
Figure 4.
Spectral analysis of some wrong classification examples. The bars are rendered in different colors according to different categories. (a) Crystal–sulfur. (b) chlorite–saponite.
Figure 4.
Spectral analysis of some wrong classification examples. The bars are rendered in different colors according to different categories. (a) Crystal–sulfur. (b) chlorite–saponite.
Figure 5.
Visualization comparison using the T-SNE technique. (a) Original spectral signatures; (b) highly discriminative features derived from the developed method.
Figure 5.
Visualization comparison using the T-SNE technique. (a) Original spectral signatures; (b) highly discriminative features derived from the developed method.
Figure 6.
Effect of the number of transformer layers on classification results.
Figure 6.
Effect of the number of transformer layers on classification results.
Figure 7.
Effect of the number of transformer heads on classification results.
Figure 7.
Effect of the number of transformer heads on classification results.
Table 1.
Characteristics of the multispectral camera onboard the Zhurong rover.
Table 1.
Characteristics of the multispectral camera onboard the Zhurong rover.
Parameters | Values |
---|
Number of channels | 9 (panchromatic and multispectral) |
Weight | 1.2 kg |
Geometric resolution | |
Radiometric resolution | 10 bits |
Pixel size | 5.5 |
Focal length | 50 mm |
Imaging distance | |
Table 2.
The statistics of image patches in training set, validation set and test set. For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr).
Table 2.
The statistics of image patches in training set, validation set and test set. For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr).
Category | Mu | He | Ca | Gal | Ka | Ta | Pyr | Cha | GT | St | Gab |
Training set | 3011 | 5508 | 3149 | 4019 | 3667 | 3486 | 3360 | 3894 | 3824 | 4272 | 6044 |
Validation set | 502 | 918 | 525 | 670 | 612 | 581 | 560 | 649 | 639 | 712 | 1008 |
Test set | 1506 | 2754 | 2575 | 2009 | 1834 | 1743 | 1680 | 1947 | 1917 | 2136 | 3022 |
Category | Qu | Chl | Se | Sm | Te | Gy | Gr | Cr | Sua | Ar | Bas |
Training set | 4350 | 4882 | 3131 | 3686 | 5186 | 3378 | 3311 | 4043 | 3272 | 3900 | 4100 |
Validation set | 725 | 814 | 522 | 614 | 864 | 563 | 552 | 674 | 545 | 650 | 683 |
Test set | 2175 | 2441 | 1566 | 1843 | 2593 | 1689 | 1655 | 2022 | 1637 | 1950 | 2050 |
Category | Fl | Ps | Sa | Go | Bar | Sul | BV | Pr | Sum | In Total |
Training set | 3840 | 3378 | 5280 | 4572 | 4050 | 4356 | 6038 | 4708 | 128,045 | 214,421 |
Validation set | 640 | 563 | 880 | 762 | 675 | 726 | 1006 | 785 | 21,344 |
Test set | 1920 | 1689 | 2640 | 2286 | 2025 | 2178 | 3020 | 2355 | 65,032 |
Table 3.
Classification results generated by the developed method in precision, recall and per class as well as overall accuracy (the unit: %). For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr).
Table 3.
Classification results generated by the developed method in precision, recall and per class as well as overall accuracy (the unit: %). For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr).
Category | Mu | He | Ca | Gal | Ka | Ta | Pyr | Cha | GT | St | Gab |
Precision | 100.0 | 96.62 | 97.07 | 93.84 | 99.89 | 100.0 | 100.0 | 100.0 | 100.0 | 92.54 | 99.93 |
Recall | 100.0 | 99.93 | 98.98 | 97.95% | 100.0 | 100.0 | 100.0 | 100.0 | 100.0 | 100.0 | 99.70 |
F1-score | 100.0 | 98.25 | 98.01 | 95.86 | 99.94 | 100.0 | 100.0 | 100.0 | 100.0 | 96.13 | 99.81 |
Category | Qu | Chl | Se | Sm | Te | Gy | Gr | Cr | Sua | Ar | Bas |
Precision | 99.95 | 89.58 | 99.23 | 100.0 | 95.22 | 100.0 | 100.0 | 81.53 | 99.51 | 100.0 | 100.0 |
Recall | 100.0 | 82.09 | 99.93 | 100.0 | 100.0 | 100.0 | 100.0 | 100.0 | 100.0 | 100.0 | 100.0 |
F1-score | 99.97 | 85.67 | 99.58 | 100.0 | 97.55 | 100.0 | 100.0 | 89.82 | 99.76 | 100.0 | 100.0 |
Category | Fl | Ps | Sa | Go | Bar | Sul | BV | Pr | Overall accuracy |
Precision | 100.0 | 99.12 | 96.35 | 100.0 | 100.0 | 99.29 | 93.34 | 88.28 | 96.92% |
Recall | 100.0 | 100.0 | 100.0 | 99.65 | 100.0 | 58.40 | 85.03 | 98.59 |
F1-score | 100.0 | 99.55 | 98.14 | 99.82 | 100.0 | 73.54 | 88.99 | 93.16 |
Table 4.
Effect of the size of image patches on classification results in per class and overall accuracy (the unit: %). For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr). The best performance is highlighted in BOLD.
Table 4.
Effect of the size of image patches on classification results in per class and overall accuracy (the unit: %). For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr). The best performance is highlighted in BOLD.
Patch Size | Mu | He | Ca | Gal | Ka | Ta | Pyr | Cha | GT | St | Gab |
Size = 5 | 95.18 | 95.86 | 92.49 | 94.19 | 98.66 | 99.97 | 100.0 | 99.80 | 99.44 | 97.45 | 98.33 |
Size = 7 | 99.94 | 97.57 | 92.23 | 95.21 | 99.65 | 100.0 | 100.0 | 100.0 | 99.93 | 97.04 | 99.31 |
Size = 9 | 100.0 | 98.25 | 98.01 | 95.86 | 99.94 | 100.0 | 100.0 | 100.0 | 100.0 | 96.13 | 99.81 |
Size = 11 | 100.0 | 97.97 | 87.88 | 97.47 | 99.88 | 100.0 | 100.0 | 99.95 | 99.73 | 98.15 | 97.77 |
Patch size | Qu | Chl | Se | Sm | Te | Gy | Gr | Cr | Sua | Ar | Bas |
Size = 5 | 99.58 | 81.66 | 98.54 | 99.66 | 91.09 | 99.91 | 99.34 | 92.01 | 99.54 | 99.83 | 99.77 |
Size = 7 | 99.56 | 84.49 | 99.56 | 100.0 | 96.81 | 99.97 | 99.27 | 89.97 | 99.70 | 99.97 | 99.97 |
Size = 9 | 99.97 | 85.67 | 99.58 | 100.0 | 97.55 | 100.0 | 100.0 | 89.82 | 99.76 | 100.0 | 100.0 |
Size = 11 | 99.33 | 86.88 | 96.55 | 99.93 | 96.99 | 100.0 | 100.0 | 99.21 | 100.0 | 99.96 | 99.94 |
Patch size | Fl | Ps | Sa | Go | Bar | Sul | BV | Pr | Overall accuracy |
Size = 5 | 100.0 | 98.88 | 97.17 | 99.32 | 99.98 | 67.59 | 85.33 | 93.13 | 95.96 |
Size = 7 | 100.0 | 99.18 | 96.87 | 99.74 | 100.0 | 67.27 | 89.58 | 92.51 | 96.48 |
Size = 9 | 100.0 | 99.55 | 98.14 | 99.82 | 100.0 | 73.54 | 88.99 | 93.16 | 96.92 |
Size = 11 | 100.0 | 99.93 | 96.36 | 100.0 | 100.0 | 68.77 | 89.33 | 87.46 | 96.83 |
Table 5.
Effect of category-aware contrastive loss on classification results in per class (%). For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr). The best performance is highlighted in BOLD.
Table 5.
Effect of category-aware contrastive loss on classification results in per class (%). For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr). The best performance is highlighted in BOLD.
Methods | Mu | He | Ca | Gal | Ka | Ta | Pyr | Cha | GT | St | Gab |
No contrastive loss | 100.0 | 97.57 | 91.21 | 95.87 | 98.84 | 100.0 | 100.0 | 100.0 | 100.0 | 96.34 | 99.46 |
Proposed method | 100.0 | 98.25 | 98.01 | 95.86 | 99.94 | 100.0 | 100.0 | 100.0 | 100.0 | 96.1 | 99.81 |
Methods | Qu | Chl | Se | Sm | Te | Gy | Gr | Cr | Sua | Ar | Bas |
No contrastive loss | 99.97 | 80.76 | 99.45 | 100.0 | 95.85 | 100.0 | 100.0 | 93.03 | 99.87 | 99.97 | 99.97 |
Proposed method | 99.97 | 85.67 | 99.58 | 100.0 | 97.55 | 100.0 | 100.0 | 89.82 | 99.76 | 100.0 | 100.0 |
Methods | Fl | Ps | Sa | Go | Bar | Sul | BV | Pr | Average | Overall accuracy |
No contrastive loss | 100.0 | 99.97 | 94.70 | 99.91 | 100.0 | 68.77 | 89.14 | 91.49 | 96.40 | 96.25 |
Proposed method | 100.0 | 99.55 | 98.14 | 99.82 | 100.0 | 73.54 | 88.99 | 93.16 | 97.12 | 96.92 |
Table 6.
Comparisons with other methods in per class and overall accuracy (%). For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr). The best performance is highlighted in BOLD fonts.
Table 6.
Comparisons with other methods in per class and overall accuracy (%). For abbreviation, muscovite (Mu), hematite (He), calcite (Ca), galena (Gal), kaolinite (Ka), talcum (Ta), pyrite (Pyr), chalcopyrite (Cha), gray tiemannite (GT), stibnite (St), gabbro (Gab), quartz (Qu), chlorite (Chl), serpentine (Se), smaltite (Sm), tennantite (Te), gypsum (Gy), graphite (Gr), crystal (Cr), suanite (Sua), aragonite (Ar), basalt (Bas), fluorite (Fl), psilomelane (Ps), saponite (Sa), goethite (Go), barite (Bar), sulfur (Sul), biotite vermiculite (BV), and pyroxenite (Pr). The best performance is highlighted in BOLD fonts.
Methods | Mu | He | Ca | Gal | Ka | Ta | Pyr | Cha | GT | St | Gab |
Decision tree [56] | 95.98 | 85.82 | 90.06 | 90.86 | 99.20 | 95.20 | 99.97 | 87.30 | 97.62 | 95.06 | 86.29 |
Random forest [57] | 97.98 | 88.83 | 95.60 | 91.39 | 99.91 | 98.44 | 100.0 | 88.82 | 99.47 | 94.72 | 90.52 |
SVM [58] | 99.70 | 88.44 | 97.99 | 95.68 | 98.86 | 98.78 | 99.97 | 88.62 | 99.63 | 99.09 | 96.17 |
ConvNet [59] | 99.33 | 91.66 | 93.26 | 94.96 | 97.39 | 99.45 | 100.0 | 87.03 | 99.76 | 97.62 | 98.10 |
Developed method | 100.0 | 98.25 | 98.01 | 95.86 | 99.94 | 100.0 | 100.0 | 100.0 | 100.0 | 96.13 | 99.81 |
Methods | Qu | Chl | Se | Sm | Te | Gy | Gr | Cr | Sua | Ar | Bas |
Decision tree [56] | 99.15 | 57.40 | 62.62 | 85.76 | 90.48 | 99.20 | 97.58 | 95.95 | 87.53 | 96.26 | 92.66 |
Random forest [57] | 99.38 | 65.13 | 70.33 | 82.94 | 91.69 | 100.0 | 99.16 | 98.92 | 93.77 | 96.27 | 96.28 |
SVM [58] | 99.70 | 59.31 | 77.72 | 72.96 | 83.12 | 100.0 | 97.66 | 97.84 | 93.84 | 99.82 | 98.77 |
ConvNet [59] | 100.0 | 74.56 | 76.19 | 99.53 | 96.26 | 100.0 | 99.84 | 98.38 | 99.31 | 99.58 | 99.72 |
Developed method | 99.97 | 85.67 | 99.58 | 100.0 | 97.55 | 100.0 | 100.0 | 89.82 | 99.76 | 100.0 | 100.0 |
Methods | Fl | Ps | Sa | Go | Bar | Sul | BV | Pr | Average | Overall accuracy |
Decision tree [56] | 98.23 | 96.93 | 86.79 | 97.42 | 98.99 | 69.38 | 80.17 | 91.37 | 90.24 | 90.22 |
Random forest [57] | 99.89 | 98.39 | 91.67 | 97.62 | 99.38 | 71.56 | 83.98 | 94.00 | 92.53 | 92.41 |
SVM | 99.94 | 95.74 | 87.25 | 98.55 | 100.0 | 30.72 | 77.18 | 90.77 | 90.79 | 90.79 |
ConvNet [59] | 99.03 | 99.59 | 95.29 | 98.55 | 98.15 | 91.88 | 79.34 | 98.32 | 95.04 | 95.24 |
Developed method | 100.0 | 99.55 | 98.14 | 99.82 | 100.0 | 73.54 | 88.99 | 93.16 | 97.12 | 96.92 |