Comparison of CNN models for optimizing Banana Image Classification

This study focused on image recognition of bananas by utilizing the concept of machine learning through several CNN models. In carrying out feature extraction processes which could support the classification process of banana images. But before that, each image will be resized so that the size of each image has the same pixel size. Furthermore, each image will experience changes in scale and rotation simultaneously. This experiment used 9 classes of banana images with a total of 7936 images where each image had a different scale and rotation. The images were divided from 7504 training datasets, and 432 testing datasets. After doing the experiment used some CNN model such as ResNet50V2, InceptionV3, NasnetMobile, VGG16, VGG 19, MobileNetV2, and EfficientNet the result became a breakthrough in identifying and categorizing banana species optimally. Based on several experiments have been proven effective in recognizing and identifying banana images with various experimental results so that the highest accuracy reaches 89% in dataset testing using the EfficientNet model and 83.8% using the VGG16 model. So the research was very successful in recognizing 9 classes of bananas. However, improvements still need to be made to get better results in others classification in the next experiment.