<![CDATA[Previous research utilizing Convolutional Neural Network (CNN)-based deep learning has successfully identified chicken embryo development through candling images. However, challenges arise in spinner-rack incubators equipped with a single candling camera, eggs in non-optimal positions receive insufficient illumination, making it difficult to recognize embryonic patterns. Adding more cameras or incandescent lights risks increasing both costs and incubation temperatures. This study aims to reduce the domain gap by optimizing preprocessing using Contrast Limited Adaptive Histogram Equalization (CLAHE) on RGB candling images. The dataset comprises 12,782 candling images 11,800 for training and 982 for evaluation labeled as fertile, infertile, or dead across days 1 through 10. The CNN model was trained using a batch size of 32 and an image size of 224×224, and was evaluated based on accuracy, precision, recall, and F1-score. Experimental results demonstrate that RGB images processed with CLAHE yielded the best performance, achieving 97.37% accuracy surpassing grayscale (96.36%), RGB without CLAHE (96.16%), and grayscale with CLAHE (93.84%). Testing on 107 external images further confirmed the superiority of the RGB + CLAHE method, which produced 103 correct predictions and only 4 errors, compared to standard RGB, which yielded 101 correct predictions and 6 errors. Remaining classification errors primarily occurred in the class of embryos that failed to develop due to the blood ring phenomenon, which visually resembles a fertile embryo at day 5.]]>
