Enhanced Fish Freshness Classification with Incremental Handcrafted Feature Fusion

Phi-Hung Hoang, Nam-Thuan Trinh, Van-Manh Tran, Thi-Thu-Hong Phan · 2025

Read Enhanced Fish Freshness Classification with Incremental Handcrafted Feature Fusion by Phi-Hung Hoang, Nam-Thuan Trinh, Van-Manh Tran, Thi-Thu-Hong Phan (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-17145.txt.

Introduction

Phi-Hung Hoang and three co-authors grade fish freshness from eye photos in this 2025 paper. Freshness sets price and health risk, yet human checks vary and lack a fixed scale. Spoilage signs in eyes are faint and differ by fish type. The team builds scores from hand-made traits of each photo.

Traits cover color spread, tone bands, and surface marks in the eye. The team adds trait groups step by step to test each one's worth. One tuned tree model beat a prior deep model by a wide margin. With more training photos, a neural net scored near ninety-seven percent right. The clear trait list helps shops check fish in a steady way.

Catalog record

CREATORPhi-Hung Hoang, Nam-Thuan Trinh, Van-Manh Tran, Thi-Thu-Hong Phan
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE61 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-17145

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Enhanced Fish Freshness Classification with Incremental Handcrafted Feature Fusion Phi-Hung Hoang, Nam-Thuan Trinh, Van-Manh Tran, Thi-Thu-Hong Phan Abstract Accurate assessment of fish freshness remains a major challenge in the food industry, with direct consequences for product quality, market value, and consumer health. Conventional sensory evaluation is inherently subjective, inconsistent, and difficult to standardize across contexts, often limited by subtle, species-dependent spoilage cues. To address these limitations, we propose a handcrafted feature–based approach that systematically extracts and incrementally fuses complementary descriptors, including color statistics, histograms across multiple color spaces, and texture features such as Local Binary Patterns (LBP) and Gray-Level Co-occurrence Matrices (GLCM), from fish eye images. Our method captures global chromatic variations from full images and localized degradations from ROI segments, fusing each independently to evaluate their effectiveness in assessing freshness. Experiments on the Freshness of the Fish Eyes (FFE) dataset demonstrate the approach’s effectiveness: in a standard train–test setting, a LightGBM classifier achieved 77.56% accuracy, a 14.35% improvement over the previous deep learning baseline of 63.21%. With augmented data, an Artificial Neural Network (ANN) reached 97.16% accuracy, surpassing the prior best of 77.3% by 19.86%. These results demonstrate that carefully engineered, handcrafted features, when strategically processed, yield a robust, interpretable, and reliable solution for automated fish freshness assessment, providing valuable insights for practical applications in food quality monitoring. 1 Introduction Fish is widely recognized as a highly nutritious food source, rich in protein and essential to human diets. However, ensuring the preservation and reliable evaluation of fish quality remains a significant practical challenge. Assessing fish freshness - a decisive factor for market value, product safety, and consumer health - is particularly critical. Conventional evaluation methods, which rely on human sensory judgment, are inherently subjective, error-prone, time-consuming, and difficult to reproduce consistently [ 1 , 2 ] . Moreover, the visual characteristics of fish, especially the eyes, undergo subtle temporal changes influenced by species, storage conditions, and image acquisition settings. These challenges make the automatic classification of fish freshness from images a complex task that demands specialized and robust methodological approaches. To address these issues, many studies have explored a range of techniques, from traditional machine learning with handcrafted features, which primarily exploit fine-grained differences in color, texture, and morphology to distinguish freshness levels [ 3 , 4 , 5 ] , to advanced deep learning models that autonomously extract discriminative representations [ 6 , 7 , 8 ] . Despite such advancements, the Freshness of the Fish Eyes (FFE) dataset, an available resource for fish freshness classification, has yielded limited results, with deep learning studies reporting accuracies as low as 63.21% using a custom MobileNetV1 architecture [ 9 ] and 77.3% with deep feature extraction combined with machine learning classifiers [ 10 ] . These limitations highlight the need for alternative approaches, particularly those leveraging handcrafted features, which remain underexplored in this context. Inspired by the proven effectiveness of handcrafted feature extraction in other agricultural domains, such as rice seed purity classification [ 11 , 12 ] and plant disease recognition [ 13 , 14 , 15 ] , we have investigated its potential on the FFE dataset. Specifically, we extract a comprehensive set of descriptors, including color-based features (color statistics, color percentiles, variance ratios, histograms) and texture-based features (LBP and GLCM), from the fish eye region. To maximize their complementary properties, we employ an incremental fusion strategy that systematically optimizes feature combinations, thereby enhancing both classification accuracy and practical applicability. The key contributions of this research are as follows: • Conducting experiments on both full images and region-of-interest (ROI) segments. • Investigating a diverse set of handcrafted features, covering color and texture descriptors. • Implementing an incremental fusion strategy to optimize feature combinations. • Re-implementing a prior study to demonstrate the superior effectiveness of handcrafted features over deep learning approaches. The paper is organized as follows: Section 2 reviews related work, Section 3 details the methodology, Section 4 presents experiments, 5 provides results and discussion, and Section 6 offers conclusions and future directions. 2 Related works Assessing fish freshness is a crucial factor in ensuring quality and safety throughout the seafood supply chain. Consequently, a substantial body of research has been devoted to automating this process. While methods based on spectroscopy and specialized chemical sensors have emerged as predominant approaches, alternative modalities such as image analysis using machine learning and deep learning have also been actively explored. Initial research efforts primarily focused on traditional machine learning algorithms combined with handcrafted feature extraction, particularly from the characteristics of the fish’s eyes and gills. Early work validated that RGB color indices from eye and gill images correlate with the decomposition process, proving especially effective after the third day [ 16 ] . Subsequent study, Huang et al. demonstrated that computer vision technology outperformed NIR spectroscopy for fish freshness classification, achieving 90.00% accuracy on the prediction set compared to 80.00% for the NIR-based model [ 17 ] . When data from both techniques were combined, the model’s accuracy increased to 93.33%. Alaimahal et al. developed a method to detect fish freshness using a watershed transform for segmentation and a wavelet transform for feature extraction [ 18 ] . The classification, performed by a K-Nearest Neighbors (K-NN) classifier, achieved a correct detection rate of 90%. More recent studies continue to demonstrate the high potential of this approach. Specifically, the efficacy of the K-Nearest Neighbors (KNN) algorithm was showcased using RGB color and GLCM texture features from fish eye images, attaining a classification accuracy of 97% [ 4 ] . In another study, the combination of HSV color features with GLCM using an SVM model achieved an accuracy of 94.28% [ 19 ] . Furthermore, Rosário developed an automatic algorithm to detect the fish eye’s pupil and extract colorfulness and brightness features, achieving 92.7% accuracy through ensembled bagged trees [ 5 ] . In recent years, the field has pivoted significantly towards deep learning, particularly Convolutional Neural Networks (CNNs), which automate both feature extraction and classification. Early success was demonstrated with the VGG-16 architecture, which achieved a remarkable accuracy of 98.21% [ 6 ] . In the same year, Mohammadi Lalabadi et al. found that features from gills (96%) were more effective than those from eyes (84%) [ 3 ] . Prasetyo et al. designed a customized deep learning architecture, termed MB-BE (a modified version of MobileNetV1), which achieved 63.21% accuracy on the FFE dataset, highlighting the challenge deep learning models face in distinguishing subtle variations in fish freshness [ 9 ] . Later, Akgül et al. developed a hybrid model that successfully detected fish freshness. Their model used YOLOv5 for object detection along with Inception-ResNet-v2 and Xception for classification. This resulted in high accuracy, with the YOLOv5 and Inception-ResNet-v2 model achieving 97.67% on anchovy and 96.00% on horse mackerel datasets [ 7 ] . To improve performance, recent studies have shifted toward hybrid systems that leverage pre-trained deep learning models for feature extraction and traditional machine learning methods for classification. For example, Kılıçarslan et al. achieved a 100% classification accuracy by leveraging deep feature extractors (Xception, MobileNetV2) alongside classical classifiers (SVM, LR) [ 8 ] . Following this approach, Yildiz et al. achieved 77.3% accuracy on the FFE dataset by combining VGG19

Read the complete text