Abstract

Sorting Multiple or single electronics components manually can be time consuming as well difficult as those minute components cannot be much visible and be differentiated with naked eye, so sorting these components require high end processing machines as well high-end capturing camera devices. In this study we have used a low cost ESP32 cam module which can capture images of 360 pixels, integrating this low-cost edge device with machine learning algorithm called as SSD Mobile Net FPN 320 V2 lite can detect the minute components as well as differentiate it based on its physical attribute. This study involves objects like Green LED, White LED, Blue LED and Resistor for sorting and deployment procedure. Firstly, utilizing the Low-cost camera we have captured 400+ images of single components for sorting, then we have annotated all the captured components with respective labels and this will be trained in the TensorFlow platform using SSD Mobile Net FPN 320 V2 lite format, thus model of annotated images will be obtained. This model is trained 40,000 times in a batch of 16 epochs which helps in training all the physical attributes of the image. Secondly, the trained model is deployed on the laptop using Python IDE. Thirdly, Initialize the real time camera for capturing real time images and create a connection between ESP32 cam module and Laptop for sharing real time data. This model utilized the WIFI host communication of ESP32 Cam module as it has a built-in WIFI module, utilizing this feature helps to live feed video from ESP32 to the PC via WIFI, Next in Python IDE the WIFI URL data of the host is included such that the Machine learning SSD Mobile Net FPN 320 V2 Lite model is implemented on the live stream video, which detects all the components by comparing it with the annotated images. The proposed model shows prominent results for detection, sorting and deploying of the images by utilizing this SSD Mobile Net FPN 320 V2 lite model and also with a low-cost edge device of 360px resolution.