Towards a general framework for accurate deep learning identification of soil mesofauna: Application and validation with a focus on oribatid mites

Abstract

Morphological identification of soil oribatid mites is constrained by taxonomic complexity and reliance on specialized expertise, limiting its practical applicability and hindering the efficiency of soil mesofaunal diversity assessments. Despite advances in deep learning-based automated classification, key process factors influencing model performance remain underexplored. This study aimed to establish a generalized convolutional neural network (CNN)-based framework for rapid, accurate identification of nine co-distributed oribatid mite species from subtropical forests (Tianmu and Guan Mountains, China). We systematically evaluated the effects of CNN architectures (AlexNet, VGG16, ResNet50, ResNet101, DenseNet), image resolutions (32×32 to 224×224 pixels), background colors (eight RGB treatments), and habitat origins on classification performance using a balanced dataset (118 images per species per habitat). DenseNet achieved the highest accuracy (99.32%) at 224×224 pixels with white backgrounds (RGB 255,255,255) and successfully distinguished habitat origins (accuracy: 92.45%–100%). A standardized workflow for image acquisition, dataset construction, and model optimization was proposed. This work bridges computer vision and soil zoology, advances Soil Animal Informatics, and offers a scalable solution for large-scale soil mesofauna biodiversity monitoring in the big data era.

Graphical abstract

Keywords

soil biodiversity / deep learning / convolutional neural networks (CNNs) / species identification / morphological traits / image classification.

Highlight

	● A generalized CNN framework identifies 9 co-distributed subtropical oribatid mites.
	● DenseNet + 224×224 + white background achieves 99.32% accuracy.
	● The model distinguishes oribatid mite habitat origins (92.45%–100% accuracy).
	● A standardized workflow advances Soil Animal Informatics and monitoring.

References

[1] Ärje, J., Melvad, C., Jeppesen, M.R., Madsen, S.A., Raitoharju, J., Rasmussen, M.S., Iosifidis, A., Tirronen, V., Gabbouj, M., Meissner, K., Høye, T.T., 2020. Automatic image-based identification and biomass estimation of invertebrates. Methods in Ecology and Evolution11, 922–931.

[2] Arribas, P., Andújar, C., Moraza, M.L., Linard, B., Emerson, B.C., Vogler, A.P., 2020. Mitochondrial metagenomics reveals the ancient origin and phylodiversity of soil mites and provides a phylogeny of the Acari. Molecular Biology and Evolution37, 683–694.

[3] Bardgett, R.D., van der Putten, W.H., 2014. Belowground biodiversity and ecosystem functioning. Nature515, 505–511.

[4] Bauer, E.A., 2022. Progressive trends on the application of artificial neural networks in animal sciences - A review. Veterinarni Medicina67, 219–230.

[5] Chen, Q.J., Ding, Y.C., Liu, C., Liu, J., He, T.T., 2021. Research on spider sex recognition from images based on deep learning. IEEE Access9, 120985–120995.

[6] Chen, X.Y., Daniell, T.J., Neilson, R., O’Flaherty, V., Griffiths, B.S., 2010. A comparison of molecular methods for monitoring soil nematodes and their use as biological indicators. European Journal of Soil Biology46, 319–324.

[7] Cordes, P., Pan, X., Murvanidze, M., Seniczak, A., Scheu, S., Schaefer, I., Maraun, M., Heimburger, B., 2024. Convergent evolution revealed by paraphyly and polyphyly of many taxa of oribatid mites: a molecular approach. Experimental and Applied Acarology93, 787–802.

[8] Dong, B.Z., Nie, L.N., Jing, W.P., Cui, H., 2018. Identification method of ambrostoma quadriimpressum motschlsky based on Faster R-CNN. Computer Engineering and Applications54, 89–93,108.

[9] Gao, M.X., Zhu, J.Q., Liu, S., Cheng, X., Liu, D., Li, Y.S., 2023. Theory, method and technique of soil animal knowledge graph construction: a case study of soil mites in Tianmu Mountain, Zhejiang Province. Acta Ecologica Sinica43, 6862–6877.

[10] Gao, Y.Y., Xue, X.B., Qin, G.Q., Li, K., Liu, J.H., Zhang, Y.L., Li, X.J., 2024. Application of machine learning in automatic image identification of insects - a review. Ecological Informatics80, 102539.

[11] Gaston, K.J., Chown, S.L., Evans, K.L., 2008. Ecogeographical rules: elements of a synthesis. Journal of Biogeography35, 483–500.

[12] Ge, X.Y., Ermilov, S.G., Kazakov, D.V., Salavatulin, V.M., Kolesnikov, V.B., Sheykin, S.D., Liu, D., 2024. A small step can also be a big step—first study on DNA barcoding in species delimitation of oribatid mites in Russia. Systematic and Applied Acarology29, 631–654.

[13] Hawkins, D.M., 2004. The problem of overfitting. Journal of Chemical Information and Computer Sciences44, 1–12.

[14] He, K.M., Zhang, X.Y., Ren, S.Q., Sun, J., 2016. Deep residual learning for image recognition. In: Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE770–778.

[15] Hu, J., Shen, L., Albanie, S., Sun, G., Wu, E.H., 2020. Squeeze-and-excitation networks. IEEE Transactions on Pattern Analysis and Machine Intelligence42, 2011–2023.

[16] Huang, G., Liu, Z., Pleiss, G., van der Maaten, L., Weinberger, K.Q., 2022. Convolutional networks with dense connectivity. IEEE Transactions on Pattern Analysis and Machine Intelligence44, 8704–8716.

[17] Huang, G., Liu, Z., van der Maaten, L., Weinberger, K.Q., 2017. Densely connected convolutional networks. In: Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE2261–2269.

[18] Huang, J.H., Liu, Y.T., Ni, H.C., Chen, B.Y., Huang, S.Y., Tsai, H.K., Li, H.F., 2021. Termite pest identification method based on deep convolution neural networks. Journal of Economic Entomology114, 2452–2459.

[19] Krantz, G.W., Walter, D.E., 2009. A Manual of Acarology. Lubbock: Texas Tech University Press.

[20] Krizhevsky, A., Sutskever, I., Hinton, G.E., 2012. ImageNet classification with deep convolutional neural networks. In: Proceedings of the 26th International Conference on Neural Information Processing Systems. Lake Tahoe: Curran Associates Inc., 1097–1105.

[21] Li, Z.W., Liu, F., Yang, W.J., Peng, S.H., Zhou, J., 2022. A survey of convolutional neural networks: analysis, applications, and prospects. IEEE Transactions on Neural Networks and Learning Systems33, 6999–7019.

[22] Lin, T.Y., Goyal, P., Girshick, R., He, K.M., Dollár, P., 2020. Focal loss for dense object detection. IEEE Transactions on Pattern Analysis and Machine Intelligence42, 318–327.

[23] Liu, D., Chen, J., 2024. Checklist of oribatid mites (Acari: Oribatida) from China. Systematic and Applied Acarology29, 891–1026.

[24] Liu, J., Wang, X.W., 2020. Tomato diseases and pests detection based on improved Yolo V3 convolutional neural network. Frontiers in Plant Science11, 898.

[25] Liu, Z.L., Zhu, D., Wang, Y.F., Zhu, Y.G., Qiao, M., 2024. Collembolans maintain a core microbiome responding to diverse soil ecosystems. Soil Ecology Letters6, 230195.

[26] Oriol, T., Pasquet, J., Cortet, J., 2024. Automatic identification of Collembola with deep learning techniques. Ecological Informatics81, 102606.

[27] Qing, X., Wang, Y.H., Lu, X.Q., Li, H.B., Wang, X., Li, H.M., Xie, X.J., 2022. NemaRec: a deep learning-based web application for nematode image identification and ecological indices calculation. European Journal of Soil Biology110, 103408.

[28] Schaefer, I., Norton, R.A., Scheu, S., Maraun, M., 2010. Arthropod colonization of land -Linking molecules and fossils in oribatid mites (Acari, Oribatida). Molecular Phylogenetics and Evolution57, 113–121.

[29] Shorten, C., Khoshgoftaar, T.M., 2019. A survey on image data augmentation for deep learning. Journal of Big Data6, 60.

[30] Simonyan, K., Zisserman, A., 2014. Very deep convolutional networks for large-scale image recognition. In: Proceedings of the 3rd International Conference on Learning Representations. San Diego: ICLR, 1556.

[31] Sys, S., Weißbach, S., Jakob, L., Gerber, S., Schneider, C., 2022. CollembolAI, a macrophotography and computer vision workflow to digitize and characterize samples of soil invertebrate communities preserved in fluid. Methods in Ecology and Evolution13, 2729–2742.

[32] Thompson, J.N., 2013. Relentless Evolution. Chicago: University of Chicago Press.

[33] Witte, A., Lange, S., Lins, C., 2024. Masked autoencoder: influence of self-supervised pretraining on object segmentation in industrial images. Industrial Artificial Intelligence2, 7.

[34] Xie, J.Y., Lu, Y.Y., Kong, W.X., Xu, S.Q., 2021. Butterfly species identification from natural environment based on improved RetinaNet. Journal of Computer Research and Development58, 1686–1704.

[35] Zeng, Z.X., Liu, J., 2020. Microscopic image segmentation method of C. elegans based on deep learning. Journal of Computer Applications40, 1453–1459.

[36] Zhang, C., Guo, Y., Li, M., 2021. Review of development and application of artificial neural network models. Computer Engineering and Applications57, 57–69.

[37] Zhu, Y., Wang, P.J., Zhuang, J.Y., Zhu, Y., Xiao, J.J., Oyang, X., 2023. Multi-mode multi-feature joint intelligent identification methods for nematodes. Applied Sciences13, 7583.

RIGHTS & PERMISSIONS

Higher Education Press