Classification of Dead Cocoons Using Convolutional Neural Networks and Machine Learning Methods

Abstract

Image recognition methods classify or categorize objects by extracting significant properties from digital images of the objects and are used in the field of agriculture for quality determination. With the development of artificial intelligence technology, deep-learning techniques and tools such as convolutional neural networks (CNNs) have been used in image recognition. Existing discrimination studies have tended to extract features from images and classify them using multivariate analysis; however, deep learning algorithms have the self-learning ability to extract the feature points themselves for each neural layer. In this study, we developed models for discriminating dead cocoons using various discriminant analysis methods, including deep learning options, to establish an automation technology for the sericulture industry. A 100 W halogen light source was used for direct irradiation onto the cocoons, and a camera was positioned at the bottom of the cocoons (of which 43.9% were dead) to obtain RGB images. We conducted discrimination analyses based on the color space using four discrimination algorithms, namely, k-nearest neighbor, support vector machine, linear discriminant analysis, and partial least squares-discriminant analysis, within deep learning models (a proposed lightweight CNN model, VGG16, ResNet50, EfficientNetB0, MobileNet, ShuffleNet, GhostNet, and ConvNext). The proposed lightweight CNN model, which consisted of six convolutional layers and two fully connected layers, showed the highest discrimination accuracy (97.66%) in the Lab color space. It was thus confirmed that it is possible to automate the discrimination of dead cocoons using digital images and deep learning techniques.

Keywords

• Convolutional neural networks
• machine learning
• discrimination
• cocoon