MyoVision: A Mobile Research Tool and NEATBoost-Attention Ensemble Framework for Real Time Chicken Breast Myopathy Detection
Chaitanya Pallerla, Siavash Mahmoudi, Dongyi Wang, [6pt] Department of Biological, Agricultural Engineering, University of Arkansas, Fayetteville, AR · 2026
Read MyoVision: A Mobile Research Tool and NEATBoost-Attention Ensemble Framework for Real Time Chicken Breast Myopathy Detection by Chaitanya Pallerla, Siavash Mahmoudi, Dongyi Wang, [6pt] Department of Biological, Agricultural Engineering, University of Arkansas, Fayetteville, AR (2026) 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-2604-13456.txt.
Introduction
Pallerla, Mahmoudi, and Wang introduce MyoVision, a phone-based tool for detecting chicken breast myopathies without cutting or costly lab imaging. Woody Breast and Spaghetti Meat harm meat quality, yet checks rely on manual judgement or costly systems. MyoVision takes 14-bit RAW transillumination images, pulls texture signs of internal defects, and sorts Normal, Woody Breast, and Spaghetti Meat with a neuroevolution-optimized ensemble. On 336 fillets it reaches about 82 percent test accuracy, matching hyperspectral systems costing far more.
Catalog record
| CREATOR | Chaitanya Pallerla, Siavash Mahmoudi, Dongyi Wang, [6pt] Department of Biological, Agricultural Engineering, University of Arkansas, Fayetteville, AR |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34384 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.13456
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
MyoVision: A Mobile Research Tool and NEATBoost-Attention Ensemble Framework for Real Time Chicken Breast Myopathy Detection Chaitanya Pallerla, Siavash Mahmoudi, Dongyi Wang, [6pt] Department of Biological, Agricultural Engineering, University of Arkansas, Fayetteville, AR Abstract Woody Breast (WB) and Spaghetti Meat (SM) myopathies significantly impact poultry meat quality, yet current detection methods rely either on subjective manual evaluation or costly laboratory-grade imaging systems. We address the problem of low-cost, non-destructive multi-class myopathy classification using consumer smartphones. MyoVision is introduced as a mobile transillumination imaging framework in which 14-bit RAW images are captured and structural texture descriptors indicative of internal tissue abnormalities are extracted. To classify three categories (Normal, Woody Breast, Spaghetti Meat), we propose a NEATBoost-Attention Ensemble model, which is a neuroevolution-optimized weighted fusion of LightGBM and attention-based MLP models. Hyperparameters are automatically discovered using NeuroEvolution of Augmenting Topologies (NEAT), eliminating manual tuning and enabling architecture diversity for small tabular datasets. On a dataset of 336 fillets collected from a commercial processing facility, our method achieves 82.4% test accuracy (F1 = 0.83), outperforming conventional machine learning and deep learning baselines and matching performance reported by hyperspectral imaging systems costing orders of magnitude more. Beyond classification performance, MyoVision establishes a reproducible mobile RGB-D acquisition pipeline for multimodal meat quality research, demonstrating that consumer-grade imaging can support scalable internal tissue assessment. 1 Introduction Woody Breast (WB) and Spaghetti Meat (SM) are structural myopathies affecting poultry muscle tissue, leading to altered texture, reduced water-holding capacity, and significant economic loss in commercial processing kuttappan2016white ; soglia2016histology ; tijare2016meat . These defects primarily manifest as internal structural abnormalities rather than visible surface features, making automated detection challenging. In industrial processing lines, both conditions are identified through manual palpation or visual inspection after fillet separation approaches that suffer from limited reproducibility, operator fatigue, and inconsistent grading across facilities. Existing automated approaches depend on laboratory-grade modalities such as hyperspectral imaging pallerla2024naswd , near-infrared spectroscopy wold2019near , or custom machine vision systems yoon2022development . While effective, these systems require specialized hardware and controlled acquisition environments that fundamentally limit scalability and deployment flexibility in real processing settings usda_nass_2023 . None provide a unified framework integrating deployable sensing, automated model optimization, and multimodal data acquisition within a single practical platform. Transillumination imaging analyzes light transmitted through tissue to reveal subsurface structural variations that are not visible in conventional surface imaging, and has demonstrated utility in detecting internal defects in fruits and other agricultural products han2024non ; blasco2003machine ; huang2015development . In this study, the same principle is applied to the classification of poultry breast conditions into three categories: Normal, Woody Breast, and Spaghetti Meat, using a smartphone-based transillumination imaging system. Rather than reconstructing volumetric optical properties as performed in advanced biomedical imaging techniques such as diffuse optical tomography (DOT) or optical coherence tomography (OCT) zacharopoulos2009dot ; bowker2024oct ; ekramirad2024oct , the proposed approach captures a 2D spatially integrated attenuation pattern produced as broadband white light propagates through the chicken fillet. This attenuation pattern encodes aggregate variations in tissue density, fibrotic hardening, and fluid redistribution within the muscle structure. Woody breast and spaghetti meat myopathies exhibit fundamentally different structural characteristics, where woody breast is associated with fibrotic hardening and fluid accumulation, while spaghetti meat is characterized by connective tissue fragmentation and muscle fiber dissociation soglia2016histology ; chatterjee2016instrumental . These structural differences generate distinct macroscopic light attenuation patterns that can be captured through transillumination imaging. To leverage these patterns, a computational pipeline is developed that integrates RAW image acquisition, gradient-based statistical features, frequency-domain texture descriptors, and ensemble machine learning models optimized through neuroevolution. This framework enables effective classification of poultry breast conditions using a low-cost imaging setup while retaining meaningful structural information relevant to muscle abnormalities. Despite recent advances in machine learning for food quality inspection, existing studies rely on conventional classifiers with predefined architectures requiring extensive manual tuning geronimo2019computer ; munozlapeira2025hyperspectral ; bergstra2012random ; snoek2012practical , which may fail to capture complex structural variations in myopathic tissue while remaining susceptible to local optima on discrete, heterogeneous feature distributions goodfellow2016deep ; pascanu2013difficulty . Neuroevolution of Augmenting Topologies (NEAT) addresses these limitations by simultaneously evolving network topology and weights from minimal structures, augmenting complexity only when beneficial stanley2002evolving ; stanley2019designing , eliminating manual architecture design and gradient-based assumptions. Although neuroevolution has shown strong potential in architecture search and reinforcement learning real2019regularized ; stanley2019designing , its application in food quality evaluation remains largely unexplored. Therefore, this study explores a neuroevolution-based ensemble learning framework to improve multi-class poultry myopathy classification using smartphone-based transillumination imaging. We propose MyoVision , a smartphone-based transillumination imaging and automated ensemble learning framework for myopathy classification from small, heterogeneous tabular feature spaces. A domain-specific pipeline extracts 16 structural descriptors quantifying gradient statistics, frequency-domain texture responses, and dense tissue characteristics. For classification, we introduce the NEATBoost-Attention Ensemble , combining LightGBM with attention-based multilayer perceptrons under NEAT-driven optimization stanley2002evolving ; stanley2019designing , with optimized probability fusion leveraging complementary decision boundaries across heterogeneous base learners. Beyond classification, MyoVision integrates 14-bit RAW capture, LiDAR-based 3D point cloud acquisition, mesh reconstruction kazhdan2013screened , SAM-based segmentation kirillov2023segment , and ChatGPT-assisted analysis openai2023gpt4 as a unified mobile research platform for standardized multimodal RGB-D dataset collection. The main contributions of this work are: • A smartphone transillumination imaging framework capturing 2D spatially-integrated attenuation patterns for internal myopathy assessment without specialized optical hardware. • A neuroevolution-optimized ensemble (NEATBoost-Attention) evolving network topology and hyperparameters via NEAT for automated model discovery on small, heterogeneous datasets. • A unified mobile research platform integrating RAW imaging, LiDAR 3D acquisition, mesh reconstruction, Auto Segmentation and AI-assisted analysis for standardized multimodal RGB-D data collection and chicken breast myopathy classification. The rest of the paper is organized as follows. Section 2 reviews related work. Section 3 describes the MyoVision architecture, feature extraction pipeline, and classification methodology. Section 4 presents classification results. Section 5 contextualizes findings and discusses limitations, and Section 6 concludes the paper. 2 Related Work Recent advances in mobile computer vision have enabled smartphone-based diagnostic tools across agricultural domains. GranoScan dainelli2024granoscan deployed smartphone imagery for detection of over 80 wheat pests and diseases, achieving 77–95% accuracy under real-world conditions. AgroAId reda2022agroaid applied EfficientNet via TensorFlow Lite for multi-class plant disease classification with regional outbreak analytics. Iftikhar et al. iftikhar2024plant and Trivedi et al. trivedi2025cropleafnet proposed CNN-based mobile apps for early disease detection, while Avci et al. avci2025lightweight , Niaz et al. niaz2025efficient , and Goklani goklani2024realtime demonstrated lightweight edge-deployable frameworks for real-time plant disease inference. While these studies confirm the viability of smartphone-based agricultural diagnostics, all are single-purpose classification tools targeting surface-visible symptoms without provision for research-grade data acquisition or multimodal sensor integration. Smartphone-based meat quality assessment remains comparatively underdeveloped, with existing studies focused on narrow, task-specific surface attribute evaluation. Menezes et al. menezes2025empowering classified beef and pork tenderness from smartphone RGB images using deep neural networks, achieving F1-scores of 76.6% and 81.5% respectively. Lin et al. lin2025smartphone distinguished artificially marbled beef from premium Wagyu using GLCM and HSV features, achieving 94% binary accuracy with Random Forest. In poultry, smartphone applications have focused exclusively on live-bird disease diagnosis: Degu et al. degu2023smartphone applied fecal imaging with YOLO v3 and ResNet50 for Coccidiosis, Salmonella, and Newcastle Disease classification. A comprehensive review by Kalita et al. kalita2025application covering