[Paper Review] Nondestructive chicken egg fertility detection using CNN-transfer learning algorithms

Shoffan Saifullah, Rafał Dreżewski+8 more|arXiv (Cornell University)|Sep 28, 2023

Animal Nutrition and Physiology4 citations

TL;DR

This study proposes a nondestructive chicken egg fertility detection method using transfer learning with pre-trained CNNs (VGG16, ResNet50, InceptionNet, MobileNet) on a dataset of 200 augmented egg images. InceptionNet achieved the highest performance, attaining 98% accuracy, 100% sensitivity, and 96% specificity in classifying fertile and non-fertile eggs, demonstrating its effectiveness in multi-scale feature extraction for precision poultry hatchery applications.

ABSTRACT

This study explored the application of CNN-Transfer Learning for nondestructive chicken egg fertility detection for precision poultry hatchery practices. Four models, VGG16, ResNet50, InceptionNet, and MobileNet, were trained and evaluated on a dataset (200 single egg images) using augmented images (rotation, flip, scale, translation, and reflection). Although the training results demonstrated that all models achieved high accuracy, indicating their ability to accurately learn and classify chicken eggs' fertility state, when evaluated on the testing set, variations in accuracy and performance were observed. InceptionNet exhibited the best overall performance, accurately classifying fertile and non-fertile eggs. It demonstrated excellent performance in both training and testing sets in all parameters of the evaluation metrics. In testing set, it achieved an accuracy of 0.98, a sensitivity of 1 for detecting fertile eggs, and a specificity of 0.96 for identifying non-fertile eggs. The higher performance is attributed to its unique architecture efficiently capturing features at different scales leading to improved accuracy and robustness. Further optimization and fine-tuning of the models might necessary to address the limitations in accurately detecting fertile and non-fertile eggs in case of other models. This study highlighted the potential of CNN-Transfer Learning for nondestructive fertility detection and emphasizes the need for further research to enhance the models' capabilities and ensure accurate classification.

Motivation & Objective

• To develop a noninvasive method for detecting chicken egg fertility using deep learning to support precision poultry hatchery practices.
• To evaluate the performance of multiple transfer learning-based CNN architectures on a limited egg image dataset.
• To identify the most effective model for distinguishing fertile from non-fertile eggs using image augmentation and transfer learning.
• To assess the robustness and generalization capability of pre-trained models on small, specialized datasets in agricultural imaging.
• To provide a foundation for scalable, automated fertility screening in commercial poultry operations.

Proposed method

• Transfer learning was applied using four pre-trained CNN models: VGG16, ResNet50, InceptionNet, and MobileNet.
• A dataset of 200 single chicken egg images was augmented using rotation, flipping, scaling, translation, and reflection to improve generalization.
• The models were fine-tuned on the augmented dataset using transfer learning to adapt features to egg fertility classification.
• Evaluation metrics included accuracy, sensitivity, and specificity to assess model performance on both training and testing sets.
• The InceptionNet architecture was selected for its ability to capture multi-scale features through its inception modules.
• Performance was compared across models using standard classification metrics on a held-out test set.

Experimental results

Research questions

• RQ1 Which pre-trained CNN architecture performs best for nondestructive fertility classification of chicken eggs using transfer learning?
• RQ2 How does data augmentation impact the generalization and accuracy of transfer learning models on a small egg image dataset?
• RQ3 To what extent can transfer learning models achieve high sensitivity and specificity in distinguishing fertile from non-fertile eggs?
• RQ4 What role does multi-scale feature extraction play in improving classification performance for egg fertility detection?
• RQ5 Can transfer learning models trained on general image data effectively generalize to specialized agricultural imaging tasks like egg fertility detection?

Key findings

• InceptionNet achieved the highest test accuracy of 98%, outperforming VGG16, ResNet50, and MobileNet on the egg fertility classification task.
• InceptionNet demonstrated perfect sensitivity (1.0) in detecting fertile eggs, indicating no false negatives in the test set.
• The model achieved a specificity of 0.96, correctly identifying 96% of non-fertile eggs as such.
• All models achieved high training accuracy, indicating effective learning, but performance varied significantly on the test set.
• The superior performance of InceptionNet is attributed to its efficient multi-scale feature extraction via inception modules.
• Further fine-tuning and optimization are recommended to improve the performance of other models on this specific task.