Hasil Pencarian  ::  Simpan CSV :: Kembali

"ABSTRACTGamma ray log is a logging tool to capture the radioactive level of a rock or formation measured in API units. This logging tool generally has a capability to differentiate between permeable and impermeable layers. Usually the impermeable layer tends to have higher radioactivity compared to the permeable one except for the feldspar bearing formation. In addition, another capability of this logging tool is ti determine the kind of clay mineral by using ratio data between Thorium and Potassium. This laboratory experiment uses Spectral Gamma Ray Equipment at LEMIGAS Routine Core Laboratory. The Quality of gamma ray log measurement is significantly affected by the speed of the conveyor belt. During the experiment, the measurement speed of 30 m/hour is the optimum speed to achieve good quality data and time efficiency with the data amount of 169 points/meter. The result of SGR measurement gives the reading on the content of Uranium, Thorium, and Potassium. The Thorium and Potassium content are compared and plotted in a Quirein graphic which was modified by Schlumberger in 1985. Using this crossplot, we can identify the presence of the Chlorite, Montmorillonite, Kaolinite, Illite, mixed with layer Feldspar, Mica, Glauconite minerals and so on. A case study conducted on Wells A1, A2, A3, and A4 indicated that the result of this crossplot was similar to the measurement using XRD."

"Pada penelitian ini berkaitan dengan penerapan kemajuan kecerdasan artifisial dengan menggunakan algoritma You Only Look Once (YOLO) dalam tugas deteksi dan segmentasi pada bidang geologi yaitu untuk identifikasi mineral dengan menggunakan data petrografi. Data yang digunakan untuk proses pelatihan model deteksi dan segmentasi berjumlah 500 gambar sayatan tipis batuan beku. 500 gambar sayatan tipis, dilakukan proses anotasi secara manual dan membagi data tersebut ke dalam set pelatihan, set validasi, dan set prediksi. Pada 3 set tersebut, jumlah kelas mineral yang teranotasi adalah 6 yaitu kelas mineral plagioklas, biotit, horblend, piroksen, alkali-feldspar, dan kuarsa. Teknik augmentaasi yang diterapkan untuk mengatasi keterbatasan dataset pada penelitian ini adalah augmentasi geometri (model 1) dan mosaik (model 2). Model dengan augmentasi mosaik, menjadikan model dengan kinerja yang baik dalam tugas deteksi dan segmentasi mineral, dikarenakan augmentasi mosaik menghasilkan 1 image patch memiliki 4 variasi gambar sayatan tipis, sehingga model tersebut memiliki nilai mAP = 82.3% sedangkan model dengan augmentasi geometri nilai mAP 67.5%. Empat kelas mineral yang memiliki nilai mAP diatas 70% pada mode pelatihan dan validasi adalah mineral plagioklas, biotit, alkali-feldspar, dan piroksen. Diharapkan dari penelitian ini dapat membantu identifikasi mineral dalam sayatan tipis dengan lebih efisien dan akurat.

---

This research is related to the application of advances in artificial intelligence using the You Only Look Once (YOLO) algorithm in detection and segmentation tasks in the field of geology, namely for mineral identification using petrographic data. The data used for the detection and segmentation model training process consisted of 500 thin section images of igneous rock. 500 thin section images were annotated manually and divided the data into a training set, validation set and prediction set. In these 3 sets, the number of annotated mineral classes is 6, namely the mineral classes plagioclase, biotite, horblend, pyroxene, alkali-feldspar, and quartz. The augmentation techniques applied to overcome the limitations of the dataset in this research are geometric augmentation (model 1) and mosaic (model 2). The model with mosaic augmentation is a model with good performance in mineral detection and segmentation tasks, because mosaic augmentation produces 1 image patch with 4 variations of thin section images, so the model has a mAP value = 82.3% while the model with geometric augmentation has a mAP value of 67.5%. The four mineral classes that have mAP values above 70% in training and validation mode are the minerals plagioclase, biotite, alkali-feldspar, and pyroxene. It is hoped that this research can help identify minerals in thin sections more efficiently and accurately."