<![CDATA[Recently, Deep learning methods with Convolutional Neural Networks (CNNs) have been widely used for image classification tasks. CNN has an unrivaled advantage in extracting discriminatory image features. However, many existing CNN-based methods are designed to go deeper and more significant with more complex layers that make them challenging to implement on mobile devices or real-time devices that use microcontrollers like raspberry pi, Arduino, and immediately. This is overcome by using a Light Convolutional Neural Network (LCNN), so it needs to experiment to test the difference in LCNN performance on a personal computer and a raspberry pi four microcontrollers with a Raspbian operating system. Experiments will be carried out using several performance measures: accuracy, F-1 score, recall, precision, and time to get performance results from deep learning. As such, the results and model architecture will confirm performance differences across individual devices and show how the model performs on resource-constrained or real-time devices. Tests show that the performance of the raspberry pi, which is a tool with limited resources, does not affect the quality of image recognition but affects the recognition processing time because the raspberry pi requires a longer processing time to perform one data or photo recognition process. This will accumulate the time required for processing many data, so it can conclude that the raspberry pi and tools with limited resources are not very practical for conducting recognition training and carrying out a recognition process that contains a lot of data or photos in one process.]]>
Copyrights © 2021
