Titre :

Improving Date Fruit Sorting with a Novel Multimodal Approach and CNNs

Auteur(s) :

Boumaraf, Ibtissam [Auteur]
University of Biskra Mohamed Khider
Djeffal, Abdelhamid [Auteur]
University of Biskra Mohamed Khider
Setta, Sarah [Auteur]
University of Biskra Mohamed Khider
Tahleb Ahmed, Abdelmalik [Auteur]
Institut d’Électronique, de Microélectronique et de Nanotechnologie - UMR 8520 [IEMN]
COMmunications NUMériques - IEMN [COMNUM - IEMN]

Titre de la revue :

International journal of advances in soft computing and its applications

Pagination :

190-206

Date de publication :

2023-11

ISSN :

2074-8523

Mot(s)-clé(s) en anglais :

Convolutional Neural Network Date Fruit Image Classification Multiscale Sorting Process Thermal Image Transfer Learning Weight scale
Convolutional Neural Network
Date Fruit
Image Classification
Multiscale Sorting Process
Thermal Image
Transfer Learning
Weight scale

Discipline(s) HAL :

Physique [physics]
Sciences de l'ingénieur [physics]

Résumé en anglais : [en]

Date fruit is a beloved and widely consumed food in the Middle East and North Africa, and its popularity is growing globally. However, sorting these fruits can be a time-consuming and labor-intensive process, particularly ...
Lire la suite >Date fruit is a beloved and widely consumed food in the Middle East and North Africa, and its popularity is growing globally. However, sorting these fruits can be a time-consuming and labor-intensive process, particularly when done manually. To address this challenge, we have proposed an innovative approach that uses multimodal data fusion and convolutional neural networks (CNNs) to efficiently classify Algerian date fruit. Our process involves capturing four RGB images of the date fruit from various angles, a thermal image, and the weight of the fruit. We create a new image where the first channel consists of a grayscale image obtained by averaging the four RGB images of the fruit. The second channel contains the thermal image, and the third channel contains the normalized weight data. The new dataset is then divided into training, validation, and testing sets. We conducted experiments using four different models: VGG16, InceptionV3, ResNet50, and Basic CNN. Our findings show that the VGG16 model achieved the highest accuracy during training, validation, and testing, with scores of 99.6%, 90.4%, and 94%, respectively. The InceptionV3 model had the lowest accuracy, while the ResNet50 and Basic CNN models had similar performances. Our results indicate that the VGG16 model is the most suitable for classifying Algerian date fruit. Our proposed approach offers a promising solution to improve efficiency and accuracy, ultimately enhancing the quality of sorted fruit and increasing its market value.Lire moins >

Langue :

Anglais

Vulgarisation :

Non

Collections :

• Institut d'Électronique, de Microélectronique et de Nanotechnologie (IEMN) - UMR 8520

Source :

Harvested from HAL