Hasil Pencarian  ::  Simpan CSV :: Kembali

"instance point cloud memungkinkan untuk melakukan segmentasi bentuk dari instance 3D yang berbeda pada kelas semantik yang sama. Penerapan segmentasi 3D pada pemodelan 3D area perkotaan dapat merangsang perkembangan lebih lanjut untuk menganalisis pemodelan 3D area perkotaan. Segmentasi instance 3D point cloud perkotaan memiliki tantangan tersendiri, sebagai contoh ukuran skala besar dan bentuk instance yang lebih beragam, dibandingkan 3D point cloud di dalam ruang. Penelitian ini mengajukan optimasi dari segmentasi instance 3D point cloud pada daerah perkotaan skala besar dengan optimasi metode pencacahan menggunakan metode pencacahan overlapping dan modifikasi bagian backbone Hierarchical Aggregation 3D Instance Segmentation (HAIS) dengan 3D U-Net Attention ASPP Sparse CNN (metode proposed). Eksperimen dan evaluasi dilakukan terhadap HAIS dan metode proposed. Berdasarkan hasil eksperimen, didapati penggunaan metode pencacahan ukuran 50 overlapping dan modifikasi backbone HAIS dengan 3D U-Net Attention ASPP Sparse CNN (dengan hasil evaluasi AP = 48.78, AP50 = 60.45 dan AP25 = 65.33) memiliki tren kenaikan performa lebih baik dibandingkan dengan metode baseline (dengan hasil evaluasi AP = 44.83, AP50 = 56.48 dan AP25 = 62.36).

---

Instance segmentation of 3D point cloud is possible to perform the segmentation of 3D object shape and to differentiate instances on similar semantic class. Urban Area's large-scale 3D point cloud instance segmentation has its own challenges, namely large-scale instance forms and is more diverse, compared to indoor 3D point clouds. This study proposed optimization of 3D point cloud instance segmentation in largescale urban areas by enhancing the patching method by using overlapping method and modifying the HAIS backbone section with 3D U-Net Attention ASPP Sparse CNN (the proposed method). The experiments and evaluations will be carried out on HAIS model with baseline method from STPLS3D and our proposed method. Based on our experimental results, was found by using patching method 50 size overlapping and modification of the HAIS backbone with 3D U-Net Attention ASPP Sparse CNN (evaluation results of AP = 48.78, AP50 = 60.45 and AP25 = 65.33) has trend to increase the performance of HAIS method which is better than the baseline method (evaluation results AP = 44.83, AP50 = 56.48 and AP25."

"Online reviews have a profound impact on the customer or newbie who want to purchase or consume the product via web 2.0 e-commerce. Online reviews contain features which form half of the analysis in opinion mining. Most of the today’s systems work on the summarization taking the average of the obtained features and their sentiments leading to structured review information. Often the context surrounding the feature is undermined which helps in clearly classifying the sentiment of the review. Web 3.0 based machine interpretable Resource Description Framework (RDF) also structures these unstructured reviews in the form of features and sentiments obtained from traditional preprocessing and extraction techniques with the context data also provided for future ontology based analysis taking support of Wordnet 2.1 lexical database for word sense disambiguation and Sentiwordnet 3.0 scores used for sentiment word extraction. Many popular RDF vocabularies are helpful in the creation of such machine process-able data. In the work to follow, this instance RDF forms the basis for creating/upgrading the (available) OWL Ontology that can be used as structured data model with rich semantics towards supervised machine learning generating sentiment categories and are validated for precise sentiments. These are sent back to the interface as corresponding {feature, sentiment} pair so that reviews are filtered clearly and helps in satisfying the feature set of the customer."

"Kanker prostat merupakan salah satu penyakit yang menjadi penyebab kematian utama di kalangan pria. Deteksi dini melalui pemindaian medis dapat membantu dalam pengobatan dan penanganan yang efektif. Namun, interpretasi dari pemindaian ini seringkali sulit dan memerlukan keahlian klinis yang tinggi oleh para ahli patologi. Selain itu keterbatasan dataset publik dengan bentuk biopsi H&E dengan anotasi level biopsy hinggal level patch yang tersedia terbatas jumlahnya sehingga menyebabkan pelatihan machine learning menjadi lebih sulit. Oleh karena itu, penelitian ini bertujuan untuk mengembangkan dataset dengan model machine learning yang dapat membantu mengimprove model machine learning pengklasifikasi kanker prostat. Model machine learning yang digunakan untuk mengembangkan dataset dalam penelitian ini adalah conditional Progressive Growing GAN (ProGleason-GAN), sebuah jenis jaringan saraf tiruan yang dapat digunakan untuk mempelajari dan menghasilkan gambar sintetis dari pemindaian prostat yang telah menunjukkan hasil yang menjanjikan dalam generasi gambar sintetis beresolusi tinggi. Dataset yang ditambahkan dengan hasil gambar sintesis ProGleason-GAN digunakan untuk melatih model klasifikasi kanker prostat yaitu Semi Supervised Learning yang di gabungkan dengan Multiple Instance Learning. Dataset yang yang berisikan dataset SICAPv2 yang ditambahkan dengan hasil augmentasi ProGleason-GAN dinamakan SICAPv2 augmented. Penulis juga mengembangkan model klasifikasi dengan penambahan batch normalization yang dimana memungkinkan setiap batch data yang diberikan ke jaringan untuk dinormalisasi terlebih dahulu sebelum diolah lebih lanjut oleh jaringan. Ketika model klasifikasi ditambahkan dengan batch normalization serta dilatih dengan SICAPv2 augmented , maka nilai accuracy yang dihasilkan sebesar 76% lebih tinggi 4% model acuan.

---

Prostate cancer is a disease that is the main cause of death among men. Early detection through medical scanning can help in effective treatment and management. However, interpretation of these scans is often difficult and requires a high degree of clinical skill by pathologists. In addition, the limited number of available public datasets in the form of H&E biopsies with biopsy level to patch level annotations makes machine learning training more difficult. Therefore, this research aims to develop a dataset with a machine learning model that can help improve machine learning models for prostate cancer classification. The machine learning model used to develop the dataset in this research is Conditional Progressive Growing GAN (ProGleason-GAN), a type of artificial neural network that can be used to learn and generate synthetic images from prostate scans which has shown promising results in the generation of high-resolution synthetic images. tall. The dataset added with the ProGleason-GAN synthetic image results is used to train a prostate cancer classification model, namely Semi Supervised Learning combined with Multiple Instance Learning. The dataset containing the SICAPv2 dataset added with the results of ProGleason-GAN augmentation is called SICAPv2 augmented. The author also developed a classification model with the addition of batch normalization, which allows each batch of data given to the network to be normalized first before being further processed by the network. When the classification model was added with batch normalization and trained with augmented SICAPv2, the resulting accuracy value was 76%, 4% higher than the reference model."