Novel transfer learning approach for detecting mango fruit type and quality assessment

TL;DR

This paper introduces a new machine learning method to classify and assess mango quality using transfer learning and Random Forest, achieving high accuracy.

Contribution

The novel IncepForestNet approach combines transfer learning with Random Forest for mango classification and quality assessment.

Findings

The proposed model achieves 99% accuracy in mango variety classification.

Random Forest outperforms other machine learning algorithms in quality assessment tasks.

The method effectively extracts spatial features from mango images for accurate evaluation.

Abstract

Mango a widely consumed tropical fruit globally, showcases an extensive array of varieties distinguished by their distinct flavours, textures and appearances. The precise classification and assessment of mango varieties play a pivotal role in ensuring effective supply chain management and meeting consumer preferences. This study introduces an innovative methodology that harnesses transfer learning and machine learning techniques for the classification and quality evaluation of mango varieties. Our approach utilizes transfer learning a potent tool in deep learning, to leverage pre-trained Inception V3 models that have been trained on image datasets. Through fine-tuning these models with a dataset comprising mango images. we extract high-level features representative of different mango varieties. We introduced a novel IncepForestNet approach for the Feature Engineering mechanism from mango fruit varieties and quality assessment. The spatial feature is extracted with IncepForestNet from images of mango fruit varieties and quality assessment data. After this process, Random Forest is used to find probabilistic features. Furthermore, we integrate various machine learning algorithms to enhance classification accuracy and evaluate quality assessment attributes associated with mangoes. Our findings underscore the efficacy of the proposed approach in accurately classifying mango varieties and assessing crucial quality attributes. Additionally, we perform a comparative analysis of different machine learning algorithms to identify the most suitable technique for mango variety classification and quality assessment tasks. Our proposed model Random Forest (RF) performs outstandingly with a 99% accuracy rate and a k-fold validation score on both mango classification and quality assessment. Overall, this study presents a robust methodology that amalgamates transfer learning and machine learning techniques to facilitate the classification and quality evaluation of mango varieties. The proposed approach holds immense potential for streamlining supply chain operations and ensuring heightened consumer satisfaction within the Agriculture and Food sector.

Figures