Detection egg types and state using image processing and deep learning

The utilization of computer vision and automation for monitoring and collecting eggs is crucial for enhancing labor productivity. Deep learning and computer vision techniques have been extensively adopted by researchers and developers in diverse fields. This study proposes an efficient and lightweight network model utilizing YOLO V8. Rather than integrating many attributes simultaneously, the majority of research concentrate on a singular attribute, such as size or kind. This thesis presents an enhanced model that considers various egg types and their distinct conditions. Our methodology comprises three primary stages: the initial stage involves image preprocessing and augmentation techniques, including image cropping; the second stage employs YOLOv8 detection algorithms; and the last stage applies deep learning approaches to enhance classification accuracy. This is succeeded by a comparative analysis of algorithm performance in detection and classification according to diverse criteria. Upon examining the literature, we found that most research in this domain has concentrated on a particular type or a restricted range of egg states (e.g., intact or broken) (e.g., classification of chicken or duck eggs). This study involved the combination of chicken eggs (both intact and broken), duck eggs(both intact and broken), and quail eggs(both intact and broken). This study preserves elevated detection accuracy while diminishing the model's parameter count and computational burden. This method reduces deployment expenses and improves its suitability for robotic platforms. This study will identify a classification model to categorize six types of eggs and evaluate the performance of the CNN model against other models: "Random Forest", "K-Nearest Neighbors (KNN)" and "ResNet50". The dataset was organized into training, validation, and test sets. The analysis of the experimental findings indicates that the Convolutional Neural Network (CNN) surpasses the other models in performance. The CNN model is distinguished by its lightweight architecture, minimal parameter count, and rapid performance The prediction capabilities of the system, including egg classification, are 99% for the classification types of eggs and 100% for the classification states of eggs after processing