Optimizing convolutional neural network architectures : An experimental evaluation of key parameter tuning in smart city building project
*Vrushali Rajaram KadamCorresponding authorvrushali.kadam@bharatividyapeeth.eduDepartment of Business AdministrationYashwantrao Mohite Institute of ManagementBharati Vidyapeeth (Deemed to be University)Karad, Maharashtra, 415539, India0000-0002-3013-2245View full profile → , Dommaraju Sireeshasirialgorithms@gmail.comDepartment of ManagementRushford Business SchoolLuzern, 6010, Switzerland0009-0001-2514-7573View full profile →
* Corresponding author · click or hover a name for details
- Received:
- 01 Dec 2025
- Published Online:
- 31 Jul 2026
- Article type:
- Research Article
- Language:
- EN
- Article no.:
- JIOS-2367
- Pages:
- 2807–2820
Abstract
The unregulated and improper monitoring procedures of urban and metropolitan cities lead to poor land usage, uncontrolled urbanization, and pose a risk to the infrastructural tolerability of city life. Effective and continuous computer vision-based artificial intelligence monitoring systems lower the susceptibility risk impacts that can be created in sustainable ecosystems. In this Paper, I am describing my research work on computer vision-based Artificial Intelligence (AI) classification algorithms to monitor and classify the aerial images in monitoring city developments for improved safety, sustainability, and operational efficiency. Computer vision is a branch of AI for interpreting images and videos for software systems. The main objective of this project is to develop a Convolutional Neural Networks (CNN) model for image classification and also optimize parameters to improve the accuracy and reduce the computer resources utilization. The solution includes datasets (aerial images of cities with multiple classes), preprocessing stages, optimizing convolutional layers, preparing feature pyramids, and building neural networks, training networks, and validation steps. The results of this project show that increasing neural network layers and the number of training iterations improves accuracy but also consumes significant resources and leads to overfitting; however, reducing the number of neural layers and training iterations results in lower accuracy and underfitting. Therefore, several experiments were conducted to tune parameters and protocols such as convolutional filter dimensions, number of filters, pooling, strides, activation functions, and network size, as well as training-level parameters like learning rate, optimizer selection, epochs, batch size, and weights adjustments. Additionally, a user-friendly interactive dashboard has been built with the Streamlit framework for live demoing of solutions to the end users. It helps users by providing customized statistical reports with analysis of types of constructions and their growth trends. By integrating aerial images, AI, and mapping techniques, the solution supports industrial organizations, government authorities, and environmental agencies in achieving their targets in digitizing city aerial maps and structured planning for initiatives in building smart cities, construction and monitoring.
Keywords
Subject Classifications
References
[1] S. Niu and J. Zhang, “Image processing based on convolution neural network,” Electronics, vol. 14, no. 23, Art. no. 4649 (Nov. 2025), doi: 10.3390/electronics14234649.
[2] Y. Akbari, N. Almaadeed, S. Al-Maadeed, and O. Elharrouss, “Applications, databases and open computer vision research from drone videos and images: A survey,” Artificial Intelligence Review, vol. 54, pp. 3887–3938 (Feb. 2021), doi: 10.1007/s10462-020-09943-1.
[3] V. Vasan, N. V. Sridharan, R. J. Balasundaram, and S. Vaithiyanathan, “Ensemble-based deep learning model for welding defect detection and classification,” Engineering Applications of Artificial Intelligence, vol. 136, Art. no. 108961 (Oct. 2024), doi: 10.1016/j.engappai.2024.108961.
[4] Z. Niu, H. Pi, D. Jing, and D. Liu, “PII-GCNet: Lightweight Multi-Modal CNN Network for Efficient Crowd Counting and Localization in UAV RGB-T Images,” Electronics, vol. 13, no. 21, Art. no. 4298 (2024), doi: 10.3390/electronics13214298.
[5] G. K. K. Leizer, “Possible areas of application of drones in waste management during rail accidents and disasters,” Interdisciplinary Description of Complex Systems, vol. 16, no. 3-A, pp. 360–368 (Sep. 2018).
[6] M.-R. Hsieh, Y.-L. Lin, and W. H. Hsu, “Drone-based object counting by spatially regularized regional proposal network,” in Proc. 2017 IEEE Int. Conf. Computer Vision (ICCV), pp. 4165-4173 (2017), doi: 10.48550/arXiv.1707.05972.
[7] Y. Chen, J. Wang, and G. Wang, “Intelligent welding defect detection model on improved R-CNN,” IETE Journal of Research, vol. 69, no. 12, pp. 9235-9244 (Mar. 2022), doi: 10.1080/03772063.2022.2040387.
[8] D. Ke, “AI-driven closed-loop extrusion-based bioprinting system: From dynamic regulation to medical translation,” Bioprinting, vol. 54, Art. no. e00459 (Apr. 2026), doi: 10.1016/j.bprint.2025.e00459.
[9] Y. Li, B. Yu, B. Wang, T. H. Lee, and M. Banu, “Online quality inspection of ultrasonic composite welding by combining artificial intelligence technologies with welding process signatures,” Materials & Design, vol. 194, Art. no. 108912 (Sep. 2020), doi: 10.1016/j.matdes.2020.108912.
[10] S. Perri, F. Spagnolo, F. Frustaci, and P. Corsonello, “Welding defects classification through a convolutional neural network,” Manufacturing Letters, vol. 35, pp. 29–32 (Dec. 2022), doi: 10.1016/j.mfglet.2022.11.006.
[11] K. V. Krishna Kishore, V. R. R. Chirra, and U. Srinivasulu Reddy, “Optimized stacked deep ensemble classifier using hyperparameter tuning for facial expression recognition,” Journal of Information and Optimization Sciences, vol. 46, no. 6, pp. 2005–2015 (Oct. 2025), doi: 10.47974/JIOS-2029.
[12] G. Abdelhady, A. Mohsen, and N. Abdelnaser, “Hybrid CNN models for suspect facial recognition system,” Journal of Information and Optimization Sciences, vol. 47, no. 2, pp. 539–551 (Feb. 2025), doi: 10.47974/JIOS-1606.
