Research on the impact of convolutional neural network architectural parameters on the accuracy and computational efficiency of crop condition classification
- 1,3 Moscow Institute of Physics and Technology (National Research University) (Dolgoprudny, Moscow Region, Russia)
- 2 Moscow Institute of Physics and Technology (National Research University) (Dolgoprudny, Moscow Region, Russia); Tomsk State University of Control Systems and Radioelectronics (Dolgoprudny, Moscow Region, Russia)
Abstract
In today's environment, the use of modern technical solutions for monitoring the condition of agricultural crops is quite relevant. The existing method of expert agronomist assessment has several limitations, including low scalability and high subjectivity. Furthermore, current research on the application of modern information technologies, specifically convolutional neural networks, faces an unresolved contradiction: either high-precision, heavyweight networks are analyzed that cannot be used on peripheral devices in the field, or lightweight networks are created without examining the influence of various factors, including architectural ones, on the accuracy and computational complexity of processing.
Objective – to develop and validate a lightweight convolutional neural network for analyzing the condition of agricultural crops, specifically for classifying plant conditions into healthy and those affected by three types of diseases.
Results Based on a grid search of architectural parameters, the optimal convolutional neural network configuration was identified: 2 convolutional blocks, 32 initial channels, no batch normalization, and a fully connected layer size of 128. Next, to evaluate the proposed convolutional neural network, a comparison was conducted with modern architectural solutions: ResNet18, EfficientNetV2-S, and ConvNeXt-Tiny. It was found that the developed neural network outperforms its counterparts in both accuracy and computational complexity.
The developed convolutional neural network represents an effective, lightweight, and applicable tool for automated monitoring of grapevine health, which can be integrated into peripheral and mobile devices.
- Received
- 02.06.2026
- Approved after review
- 22.06.2026
- Accepted
- 01.09.2026
How to cite
Самарин С.В., Катаев М.Ю., Дьяченко Р.А. Исследование влияния архитектурных параметров сверточной нейронной сети на точность и вычислительную эффективность классификации состояния сельскохозяйственных культур // Информационно-измерительные и управляющие системы. 2026. Т. 24. № 5. С. 129−140. DOI: https://doi.org/10.18127/j20700814-202605-16
References
- Stukalov S.B., Kostenkov V.A. Podhody k postroeniyu algoritmov obrabotki vizual'noj informacii dlya avtomatizirovannoj orientacii vozdushnogo sudna na letnom pole/ Aktual'nye problemy i perspektivy razvitiya grazhdanskoj aviacii: materialy XIII Mezhdunar. nauch.-prakt. konf., Irkutsk, 10–11 oktyabrya 2024 goda. Irkutsk: Moskovskij gos. tekhn. un-t grazhdanskoj aviacii. 2024. S. 215–221 (In Russian).
- D'yachenko R.A., Gura D.A., Bespyatchuk D.A. i dr. K voprosu o primenenii nejrosetevogo algoritma dlya monitoringa sostoyaniya gorodskoj sredy/ Ekologicheskie sistemy i pribory. 2023. № 8. S. 42–50 (In Russian).
- Kataev M.Yu., Pas'ko O.A., Kartashov E.Yu. Analiz prakticheskih vozmozhnostej primeneniya bespilotnyh letatel'nyh apparatov v sel'skom hozyajstve/ Vestnik KrasGAU. 2023. № 1(190). S. 54-62. DOI 10.36718/1819-4036-2023-1-54-62 (In Russian).
- Samarin S.V. Sravnitel'nyj analiz metodov sbora informacii o sostoyanii vinogradnyh list'ev/ Tekhnicheskie i tekhnologicheskie sistemy: Materialy XVI Mezhdunar. nauch. konf. V 2 tomah, Krasnodar. 23–24 dekabrya 2025 goda. Krasnodar: OOO «Izdatel'skij Dom-Yug». 2025. S. 174–177 (In Russian).
- Ivanov V.A., Maksimov V.V., Galev K.D. Graficheskie uskoriteli kak sredstvo obucheniya nejronnyh setej sinteza signalov razlichnyh fizicheskih principov dejstviya/ Radiotekhnika. 2021. T. 85, № 2. S. 27-32. DOI 10.18127/j00338486-202102-04 (In Russian).
- Suri babu Nuthalapati, Mathew Midhun P., Abubeker K. M., Shafeena Karim A. SwinGNet: A Hybrid Swin Transform- GoogleNet Framework for Real-Time Grape Leaf Disease Classification, Procedia Computer Science, 2025. V. 258. P. 1629–1639. https://doi.org/10.1016/j.procs.2025.04.394
- Guoying Miao, Tianyou Liu, Yulong Zhang, Yunping Liu. An improved lightweight grape leaf disease detection algorithm based on deep learning. Digital Signal Processing. 2026. V. 179. P. 106179. https://doi.org/10.1016/j.dsp.2026.106179
- Loganathan G., Palanivelan M. An explainable lightweight convolutional neural network with global–local context-enhanced channel attention for grape leaf disease detection, Computers and Electrical Engineering. 2025. V. 128. Part B. P. 10745. https://doi.org/10.1016/j.compeleceng.2025.110745.
- Dharrao M., Zade N., Kamatchi R., Sonawane R., Henry R., Dharrao D. Grapes leaf disease dataset for precision agriculture, Data in Brief, 2025. V. 61. P. 111716. https://doi.org/10.1016/j.dib.2025.111716.
- Fareeha Naveed, Adven Masih, Jabar Mahmood, Moeez Ahmed, Aitizaz Ali, Aysha Saddiqa. Mohamed Shabbir Hamza Abdulnabi, Ebenezer Agbozo, Sustainable AI for plant disease classification using ResNet18 in few-shot learning, Array. 2025. V. 26. P. 100395. https://doi.org/10.1016/j.array.2025.100395
- Chandrasekar Venkatachalam, Priyanka Shah, Balajee A, Karthikc Raghunath K. M., Yogesh Kumaran S., Aishwarya Roy, Advanced Grape Leaf Disease Diagnosis Using EfficientNetV2L with Data Augmentation and Grad-CAM Visualization in Precision Agriculture, Procedia Computer Science. 2025. V. 260. P. 332–340. https://doi.org/10.1016/j.procs.2025.03.209.
- Banu Ulu, Seda Günaydın, Necati Çetin, Comparison performance of the CNN-based deep learning models for the distinguishing ultrasound pretreated and microwave dried jujube fruits, Measurement. 2025. V. 249. P. 117047. https://doi.org/10.1016/j.measurement. 2025.117047