Abstract:This study addresses the classification of defects in apples as a crucial measure to mitigate economic losses and optimize the food supply chain. An innovative approach is employed that integrates images from the visible spectrum and 660 nm spectral wavelength to enhance accuracy and efficiency in defect classification. The methodology is based on the use of Single-Input and Multi-Inputs convolutional neural networks (CNNs) to validate the proposed strategies. Steps include image acquisition and preprocessing, classification model training, and performance evaluation. Results demonstrate that defect classification using the 660 nm spectral wavelength reveals details not visible in the entire visible spectrum. It is seen that the use of the appropriate spectral range in the classification process is slightly superior to the entire visible spectrum. The MobileNetV1 model achieves an accuracy of 98.80\% on the validation dataset versus the 98.26\% achieved using the entire visible spectrum. Conclusions highlight the potential to enhance the method by capturing images with specific spectral ranges using filters, enabling more effective network training for classification task. These improvements could further enhance the system's capability to identify and classify defects in apples.

What problem does this paper attempt to address?

This paper attempts to solve the problem of apple defect classification in order to reduce economic losses and optimize the food supply chain. Specifically, by combining visible - light spectroscopy and 660 - nm narrow - band imaging technology, the author aims to improve the accuracy and efficiency of apple defect classification. The following are the main problems and solutions in the paper: ### Research Background and Problems 1. **Quality Problems**: Ensuring the quality of agricultural products is crucial for companies and consumers. International quality standards require high - quality fruits, vegetables, and other agricultural products to maintain human health. 2. **Cost and Market Access**: Product quality directly affects its price, and high - quality products usually sell at a higher price. Therefore, producers need an accurate system to test product quality and classify it into appropriate market levels to achieve the best economic benefits. 3. **Limitations of Manual Detection**: Currently, the agricultural industry mainly relies on manual methods to check fruit quality. This method is slow, has low precision, and is prone to missing defects that may lead to fruit rejection. ### Solutions 1. **Dataset Construction**: - Collected images of three types of apple defects (bruises, stains, and rots), including images in the visible - light spectrum and at 660 - nm wavelength. - Generated contour images for each defect type and made these images publicly available on GitHub for the scientific research community to use. 2. **Deep - Learning Model Evaluation**: - Used single - input (Single - Input) and multi - input (Multi - Inputs) convolutional neural network (CNN) architectures for defect classification. - Compared the effects of different spectral ranges (visible - light spectrum and 660 - nm wavelength) on classification performance. - Conducted experiments using pre - trained models such as MobileNetV1, DenseNet121, VGG19, and ResNet50. ### Experimental Results - **Single - Input Model**: The validation - set accuracy of MobileNetV1 on the 660 - nm wavelength dataset reached 98.80%, while the accuracy on the entire visible - light spectrum dataset was 98.26%. - **Multi - Input Model**: In the multi - input model combining the 660 - nm wavelength and visible - light spectrum datasets, MobileNetV1 achieved an accuracy of 90.91%, while the performance of other models decreased. ### Conclusions - The spectral range of 660 - nm wavelength shows slightly better performance than the entire visible - light spectrum in apple defect classification. - The multi - input model did not significantly improve the classification accuracy, probably because of the inaccuracy of the segmentation masks. - Future research is recommended to further subdivide the spectral range, including introducing the near - infrared (NIR) spectrum, to evaluate whether the model performance can be improved. Through these methods, the research shows the potential of specific spectral ranges in improving apple defect classification, which is helpful for more efficient post - production quality control.

Enhancing Apple's Defect Classification: Insights from Visible Spectrum and Narrow Spectral Band Imaging

Enhancing model robustness through different optimization methods and 1-D CNN to eliminate the variations in size and detection position for apple SSC determination

Improving Discrimination Accuracy of Pest-Infested Crabapples Using Vis/NIR Spectral Morphological Features

Multi-Band-Image Based Detection of Apple Surface Defect Using Machine Vision and Deep Learning

Nondestructive Detection of Codling Moth Infestation in Apples Using Pixel-Based NIR Hyperspectral Imaging with Machine Learning and Feature Selection

Classifying Healthy and Defective Fruits with a Multi-Input Architecture and CNN Models

Apple Defect Detection in Complex Environments

Visualized pattern recognition optimization for apple mechanical damage by laser relaxation spectroscopy

A New Simple Vision Algorithm for Detecting the Enzymic Browning Defects in Golden Delicious Apples

Feature Reduction for the Classification of Bruise Damage to Apple Fruit Using a Contactless FT-NIR Spectroscopy with Machine Learning

Apple scab detection in orchards using deep learning on colour and multispectral images

Apple Surface Defect Detection Method Based on Weight Comparison Transfer Learning with MobileNetV3

Fruit Deformity Classification through Single-Input and Multi-Input Architectures based on CNN Models using Real and Synthetic Images

Quasi Real-Time Apple Defect Segmentation Using Deep Learning

Apple Grading Method Based on Features of Color and Defect

Real-Time Grading of Defect Apples Using Semantic Segmentation Combination with a Pruned YOLO V4 Network

Identification of Defective Maize Seeds Using Hyperspectral Imaging Combined with Deep Learning

Computer Vision Detection of Defective Apples Using Automatic Lightness Correction and Weighted Rvm Classifier

Apple Surface Defect Detection Based on Gray Level Co-Occurrence Matrix and Retinex Image Enhancement

Pears Classification by Identifying Internal Defects Based on X-ray Images and Neural Networks