Quality Grading of Dried Morels Based on Improved ResNet18 and Deeply Supervised Learning

Abstract Accurate quality grading of dried morels is crucial for product value addition in the edible fungi industry; however, existing methods, including manual sorting and traditional mechanical screening, frequently encounter challenges such as low efficiency, high subjectivity, and limited capability for detecting fine surface defects. To address the challenges associated with morphological irregularities and high inter-class similarity, an improved ResNet18-based grading network, termed MSDA-ESE-ResNet18, is proposed. Specifically, the Progressive Multi-Scale Dilated Attention (MSDA) module is introduced to aggregate local defect information and global morphological fea-tures across multiple receptive fields. Meanwhile, an Enhanced Squeeze-and-Excitation (ESE) mecha-nism is employed to recalibrate channel-wise feature responses using a dual-pooling strategy. Com-bined with an auxiliary classifier for deep supervision, these modules enhance multi-scale feature rep-resentation and training stability. Experimental evaluations were conducted on a custom dataset com-prising 1805 dried morel images, with weighted random sampling and online data augmentation ap-plied to alleviate class imbalance and small-sample overfitting. Across three independent runs with different random seeds, the proposed MSDA-ESE-ResNet18 achieved an average accuracy of 95.04 ± 0.25% and a macro F1-score of 95.47 ± 0.51%, outperforming the baseline ResNet18 by 9.71 percent-age points in accuracy. The proposed model also outperformed the evaluated MobileNetV3, Effi-cientNet_b0, and ViT-B/16 models while introducing only modest computational overhead relative to ResNet18. Grad-CAM visualization indicates that the model tends to focus on discriminative mor-phological regions relevant to dried morel grading. These results demonstrate the effectiveness and computational efficiency of MSDA-ESE-ResNet18 under controlled imaging conditions and indicate its potential for future intelligent grading applications in the edible fungi processing industry. Keywords: Dried Morels, Intelligent Grading, ResNet18, Multi-Scale Dilated Attention, En-hanced Squeeze-and-Excitation

Authors

Institutions

Publication Details

Journal
Engineering Research Express
Published
2026-11-17
DOI
https://doi.org/10.1088/2631-8695/aec36d
Primary Topic
Smart Agriculture and AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Quality Grading of Dried Morels Based on Improved ResNet18 and Deeply Supervised Learning

Bin Li, Dandan Fu, Zhigang Hu, Hao Song et al.
4 links
Engineering Research Express
Smart Agriculture and AI
article

Quality Grading of Dried Morels Based on Improved ResNet18 and Deeply Supervised Learning

Bin Li, Dandan Fu, Zhigang Hu, Hao Song, junjie zuo, Quan Zhou
article en

Abstract

Abstract Accurate quality grading of dried morels is crucial for product value addition in the edible fungi industry; however, existing methods, including manual sorting and traditional mechanical screening, frequently encounter challenges such as low efficiency, high subjectivity, and limited capability for detecting fine surface defects. To address the challenges associated with morphological irregularities and high inter-class similarity, an improved ResNet18-based grading network, termed MSDA-ESE-ResNet18, is proposed. Specifically, the Progressive Multi-Scale Dilated Attention (MSDA) module is introduced to aggregate local defect information and global morphological fea-tures across multiple receptive fields. Meanwhile, an Enhanced Squeeze-and-Excitation (ESE) mecha-nism is employed to recalibrate channel-wise feature responses using a dual-pooling strategy. Com-bined with an auxiliary classifier for deep supervision, these modules enhance multi-scale feature rep-resentation and training stability. Experimental evaluations were conducted on a custom dataset com-prising 1805 dried morel images, with weighted random sampling and online data augmentation ap-plied to alleviate class imbalance and small-sample overfitting. Across three independent runs with different random seeds, the proposed MSDA-ESE-ResNet18 achieved an average accuracy of 95.04 ± 0.25% and a macro F1-score of 95.47 ± 0.51%, outperforming the baseline ResNet18 by 9.71 percent-age points in accuracy. The proposed model also outperformed the evaluated MobileNetV3, Effi-cientNet_b0, and ViT-B/16 models while introducing only modest computational overhead relative to ResNet18. Grad-CAM visualization indicates that the model tends to focus on discriminative mor-phological regions relevant to dried morel grading. These results demonstrate the effectiveness and computational efficiency of MSDA-ESE-ResNet18 under controlled imaging conditions and indicate its potential for future intelligent grading applications in the edible fungi processing industry. Keywords: Dried Morels, Intelligent Grading, ResNet18, Multi-Scale Dilated Attention, En-hanced Squeeze-and-Excitation

Engineering Research Express
Wuhan Polytechnic University (CN)
Zero hunger
Smart Agriculture and AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon