Related Experiment Video
Updated: Sep 6, 2025
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
9.3K
An Efficient Automatic Fruit-360 Image Identification and Recognition Using a Novel Modified Cascaded-ANFIS Algorithm
Namal Rathnayake1, Upaka Rathnayake2, Tuan Linh Dang3
1School of Systems Engineering, Kochi University of Technology, 185 Miyanokuchi, Tosayamada, Kami 782-8502, Kochi, Japan.
Sensors (Basel, Switzerland)
|June 24, 2022
Summary
This study introduces a novel Cascaded Adaptive Network-based Fuzzy Inference System (Cascaded-ANFIS) for comprehensive automated fruit identification. The new algorithm achieves 98.36% accuracy on the full Fruit-360 dataset, outperforming existing methods.
Area of Science:
- Computer Vision
- Machine Learning
- Artificial Intelligence
Background:
- Automated fruit identification is complex due to variations in fruit types and subtypes, often location-dependent.
- Previous studies using Convolutional Neural Network (CNN) algorithms like VGG16, Inception V3, MobileNet, and ResNet18 have not addressed the entire Fruit-360 dataset of 131 classes.
- Existing CNN models often lack computational efficiency for comprehensive fruit classification tasks.
Purpose of the Study:
- To present a novel, robust, and comprehensive study for identifying and predicting all 131 fruit classes within the Fruit-360 dataset.
- To address the limitations of previous studies by utilizing a complete dataset and improving computational efficiency.
- To introduce and evaluate the effectiveness of the Cascaded Adaptive Network-based Fuzzy Inference System (Cascaded-ANFIS) for fruit identification.
Main Methods:
- Utilized the complete Fruit-360 dataset comprising 90,483 sample images across 131 fruit classes.
- Employed a Cascaded Adaptive Network-based Fuzzy Inference System (Cascaded-ANFIS) as the core identification algorithm.
- Integrated multiple feature descriptors including Color Structure, Region Shape, Edge Histogram, Column Layout, Gray-Level Co-Occurrence Matrix, Scale-Invariant Feature Transform, Speeded Up Robust Features, Histogram of Oriented Gradients, and Oriented FAST and rotated BRIEF features.
Main Results:
- Achieved a relative accuracy of 98.36% on the comprehensive Fruit-360 dataset.
- Calculated weighted precision, recall, and F-score as 0.9843, 0.9841, and 0.9840, respectively, addressing the dataset's imbalance.
- Demonstrated superior performance and high computational efficiency compared to state-of-the-art algorithms in comparative studies.
Conclusions:
- The proposed Cascaded-ANFIS algorithm effectively handles the entire Fruit-360 dataset, offering a robust solution for automated fruit identification.
- The developed system provides high accuracy and computational efficiency, surpassing existing methods.
- This study establishes a new benchmark for comprehensive fruit classification using advanced fuzzy inference systems.
