Abstract:Kemajuan teknologi deep learning seringkali berbanding lurus dengan keterkaitan metode yang dapat diandalkan dalam penggunaan jumlah data yang besar. Convolutional Neural Network (CNN) adalah salah satu algoritma deep learning yang paling popular saat ini guna pengolahan citra. Pada era deep learning model CNN yang kompleks seperti saat ini memiliki tantangan-tantangan yang baru baik gradient vanishing, overfitting yang dikarenakan keterbatasan dataset, optimasi parameter hingga keterbatasan perangkat keras. Penelitian ini bertujuan mengukur pengaruh teknik fine tuning dan augmentasi dataset pada model transfer learning CNN Mobilenet, Efficientnet, dan Nasnetmobile dengan dataset yang variasi jumlah dataset yang memiliki jumlah yang terbatas. Pada hasil dari penelitian ini, dari ketiga dataset yang digunakan sebagai dalam melakukan training pada model efisien transfer learning baik MobileNet, EfficientNet, dan NasNetmobile, teknik augmentasi zoom range ataupun random erase dapat meningkatkan akurasi pada dataset dengan jumlah 56 citra dan 222 citra, sedangkan pada dataset dengan jumlah 500 data citra, semua teknik augmentasi terbukti dapat meningkatkan akurasi pada model arsitektur MobileNetV2 dan NasNetMobile. Sedangkan teknik fine tuning terbukti efektif dalam meningkatkan akurasi pada semua skala data yang kecil. Abstract Today deep learning technology is often associated with reliable processes (methods) when we have large amounts of data. In deep learning CNN (Convolutional Neural Network) plays a very important role which is often used to analyze (classify or recognize) visual images. In the era of deep learning models such as the complex Convolutional Neural Network (CNN) as it is today, it has new challenges such as gradient vanishing, overfitting due to dataset limitations, parameter optimization to hardware limitations. The MobileNet architecture was coined in 2017 by Howards, et al, which is one of the convolutional neural networks (CNN) architectures that can be used to overcome the need for excessive computing resources. This study aims to measure the effect of fine tune and dataset augmentation techniques on CNN mobilenet, efficientnet, and nasnetmobile transfer learning models with very small datasets. The results of this study are that of the three datasets used as the basis for training in efficient transfer learning models (mobilenet, efficientnet, and nasnetmobile), random erase and zoom range augmentation techniques dominate the increase in model accuracy. The amount of increase in accuracy after random erase or zoom range augmentation that occurs is about 0.03% to 0.1%.

Accuracy Improvement of CNN MobileNet-V1 and Residual Network 50 Layers Models using Adam setting for Car Type Classification

Study on Vehicle Recognition Algorithm Based on Convolutional Neural Network

Multiclass objects detection algorithm using DarkNet-53 and DenseNet for intelligent vehicles

The Design of Preventive Automated Driving Systems Based on Convolutional Neural Network

A new CNN-BASED object detection system for autonomous mobile robots based on real-world vehicle datasets

Vehicle classification using ResNets, localisation and spatially-weighted pooling

Augmentasi Data Pengenalan Citra Mobil Menggunakan Pendekatan Random Crop, Rotate, dan Mixup

Analisis Efek Augmentasi Dataset dan Fine Tune pada Algoritma Pre-Trained Convolutional Neural Network (CNN)

Fine-grained vehicle type classification using lightweight convolutional neural network with feature optimization and joint learning strategy

Vehicle Type Recognition Algorithm Based on Improved Network in Network

A Car Model Identification System for Streamlining the Automobile Sales Process

Improving Image Classification Robustness Using Predictive Data Augmentation

Application of an ensemble CatBoost model over complex dataset for vehicle classification

Study on the CNN model optimization for household garbage classification based on machine learning

Convolutional Neural Network Based Vehicle Classification in Adverse Illuminous Conditions for Intelligent Transportation Systems

Vehicle-Type Recognition Method for Images Based on Improved Faster R-CNN Model

Incremental Learning of Deep Neural Network for Robust Vehicle Classification

Car make and model recognition using convolutional neural network: fine-tune AlexNet architecture

SMART OPTIMIZER SELECTION TECHNIQUE: A COMPARATIVE STUDY OF MODIFIED DENSNET201 WITH OTHER DEEP LEARNING MODELS

Classification of Typical Static Objects in Road Scenes Based on LO-Net

Car Type Recognition with Deep Neural Networks