Abstract
Brand typicality is crucial in shaping consumer perceptions of brands and poses challenges for novice designers to capture due to their limited tacit knowledge. Using Weka’s image classification, we developed a brand product classification model. A dataset with 600 images was obtained from Asus and MSI, the leading eSports brands, covering various products such as controllers, mouse devices, headsets, and PC gaming components. The random forest classifier achieved an accuracy of 81 to 85%, slightly higher in the PC gaming category. The design features from Asus ROG and MSI game series products were extracted to generate 36 test images. We used keywords as prompts in Midjurney and Stable Diffusion to generate 36 test images. The developed brand product classification model in this study correctly classified 30 images. However, in the OP category, two graphics card images and one casing image were misclassified. In the PC category, two mouse images and a laptop picture were misclassified. Discrepancies between AI-generated images and personal expertise were improved in terms of the efficiency of the model for new designers. The developed model deepens the understanding of brand characteristics, maintains brand coherence, and strengthens product design innovation and market competitiveness. The model effectively assesses brand characteristics in product appearances using AI, highlighting its role in improving early design processes and new product development strategies.
1. Introduction
In the competitive eSports market, product design must blend technological innovation with brand identity to stay competitive. Brand image significantly influences consumer decisions, with product design playing a crucial role in shaping this image [1]. Moreover, deep learning and artificial intelligence (AI) technologies are increasingly used in product design, particularly for replicating brand styles [2]. We examined the application of deep learning and image recognition in eSports product design of Asus and MSI to create an aided design system aimed at enhancing design efficiency and fostering innovation.
With the continuous advancement of deep learning and AI technology, these tools are applied to product design, especially in understanding and replicating brand style. Furthermore, many product companies are leveraging AI-generated features for initial proposals or as a source of inspiration for ideas. The application of such technology in design helps novice designers who have a limited understanding of the typical characteristics of the brand better understand and apply these tools and promote product design innovation and efficiency which is the key to driving market competitiveness [3].
We established an image classification learning model based on Weka which is used as an auxiliary tool for novice designers in the initial design. This system helps them evaluate whether their designs match the typical characteristics of the brand, thereby improving the quality and innovation of their design proposals. The system uses AI-generated images as test cases, which are generated based on big data containing uncertain data sources and features. By using products available on the market as training data, the objectivity and accuracy of the assessment can be ensured.
Experts were invited to classify these images and verify the effectiveness of the system in this study. They also evaluated whether the AI-generated image classification results matched Weka’s classification results. The developed model was authenticated from the perspective of a professional designer. It improves the productivity of novice designers and enhances their accuracy in judging AI-generated images. The model reduces the need for novice and experienced designers in the early stages of design.
2. Literature Review
In the literature review, three key areas were identified. We investigated the link between brand identity and product design, emphasizing how design enhances brand image. We also explored the use of deep learning and image recognition in product design, showing how these technologies help designers replicate brand styles. AI-generated images were classified to illustrate the efficacy of these technologies in automatically assessing design schemes.
2.1. Brand Identity and Product Design
Brand image in eSports product design transcends logos, infusing design aspects with visual style and user experience. Schmitt posited that design fundamentally shapes brand strategy [4], a concept exemplified by Apple’s minimalist esthetics and focus on user-centric features. Kapferer emphasized visual representation’s role in establishing a durable brand image [5]. Balancing performance and esthetics is crucial in gaming hardware, targeting professional gamers and a wider audience. We examined how deep learning enhances product design by aligning brand identity with innovative, consistent design approaches.
2.2. Deep Learning and Image Recognition
Advances in deep learning, particularly using convolutional neural networks (CNNs), transform product design by enhancing image classification. These networks excel in handling complex image datasets, facilitating designers in capturing and replicating brand-specific styles for gaming hardware designs [6]. CNNs support the consistent application of design languages and foster innovation by generating new design possibilities [7,8]. The potential of deep learning was explored to elevate innovation and strengthen market competitiveness in eSports product design [9].
2.3. AI-Assisted Product Design
AI has been a crucial aid in product development, performing precise visual analysis and generating brand-specific design sketches. Recent advances in generative AI technology have allowed for the realism of created artifacts, making AI a valuable source of inspiration and a creative partner [10,11]. These advancements are new tools for designers to enhance creativity and efficiency. AI-generated imagery and design concepts significantly boost designers’ productivity and creativity in creating customized, brand-related designs [8]. AI-assisted designs help brands discover trends and stay market relevant by accurately incorporating brand elements [12]. In this study, three AI image generation platforms, ChatGPT 4.0 (DALL-E 2), Midjourney, and Stable Diffusion were compared to evaluate their potential and technical features in design education.
- ChatGPT 4.0 (DALL-E 2) excels at generating high-resolution, detailed images from textual descriptions, supporting educational processes by illustrating concepts and styles [13].
- Midjourney focuses on artistically inspired images with powerful stylized expressive capabilities, ideal for creative visual expression [14].
- Stable Diffusion uses diffusion model technology to fine-tune styles while maintaining image quality, making it suitable for generating diverse high-quality images, crucial for visual concepts and projects [15].
Each platform’s strengths and limitations guide selecting the appropriate AI technology in eSports hardware design. Such scientific assessment enhances design innovation and improves market competitiveness [16].
2.4. Case Selection: Asus and MSI
In the competitive global gaming hardware market, Asus and MSI impact the gaming community with their innovative designs and strong brand images. Asus targets high-end gamers with advanced technological products, while MSI emphasizes high performance and reliability for a stable gaming experience. By encroaching on Asus’s market share, MSI showcases the intense rivalry with other brands [17]. Figure 1 and Figure 2 highlight the distinct product designs and strategies of Asus and MSI, providing insights into their marketing approaches and future development strategies.
Figure 1.
Asus gaming hardware products.
Figure 2.
MSI gaming hardware products.
3. Research Procedure
In this study, we designed a Weka-based image classification system to assist novice designers in accurately grasping brand characteristics at the proposal stage. The research methodology consists of a series of steps, as displayed in Figure 3.
Figure 3.
Research procedure.
3.1. Data Collection and Preprocessing
eSports products are divided into two categories: OP and PC gaming, each with unique design features. The OP gaming category includes routers, cases, monitors, and graphics cards, with 300 images collected for each (Figure 4). The PC gaming category includes handlebars, headphones, laptops, and mice, each with 300 images (Figure 5). Figure 4 and Figure 5 display a selection, with the remaining images stored in the cloud. The dataset used in this study comprised 1200 images from Asus and MSI and analyzed for the visual characteristics of gaming hardware to enhance design efficiency and innovation. By adjusting the image resolution to 70–100 DPI and size to 700 × 700 pixels, the model effectively identified key visual features without excessive computational load. The model refined eSports product design details and optimized the design process.
Figure 4.
Asus and MSI gaming OP.
Figure 5.
Asus and MSI gaming PC.
3.2. Model Training and Testing
3.2.1. Training Process
We evaluated classifiers and the combination of BinaryPatternsPyramidFilter and RandomForest classifier. Accuracies ranged from 81 to 85% on pre-processed eSports product images. The classifiers outperformed other classifiers, such as the SimpleColorHistogramFilter paired with SMO, which yielded the lowest accuracy. The superior performance of BinaryPatternsPyramidFilter and RandomForest is attributed to their ability to handle complex visual data with texture features and ensemble learning, unlike SMO which relies solely on color features. We used 60% of the images for training. In testing, the model’s robustness and consistency were verified with an accuracy of 81% in the OP gaming category and 82% in the PC gaming category. Figure 6 and Figure 7 show the details of the training and testing. These findings demonstrate the feasibility of deep learning for eSports hardware design in model optimization and enhanced classification efficiency and accuracy.
Figure 6.
Training for OP gaming.
Figure 7.
Training for PC gaming.
3.2.2. Keyword Preparation
We collected product data from Asus ROG and MSI on websites, news, and reviews to refine keywords for an AI image generation platform. Key terms including “thin and stylish” and “high-performance GPU” were used to ensure that they accurately reflected product features (Table 1). We conducted a preliminary test on the AI platform to assess how well it interpreted and generated images, making necessary adjustments to enhance accuracy and relevance. This method boosts the experiment’s validity and refines image evaluation and classification.
Table 1.
Keywords used in This Study.
3.2.3. Image Generation
In image generation, we used such keywords as “ergonomic design”, “customizable RGB lighting”, and “high-performance GPU” to create images that reflect the core design and functional features of Asus and MSI gaming products. To ensure authenticity, we edited brand logos using image editing software to calculate the accuracy of the AI-generated images using 36 images. Table 2 illustrates the generated images of each product category.
Table 2.
Generated Images.
3.2.4. Machine Learning Classification Results
The results of image classification using Weka showed that two graphics cards and one case were misclassified in the OP category, while two mice and one laptop were misclassified in the PC category (Table 3).
Table 3.
Weka Classification Error Table.
3.2.5. Analysis of WEKA Experimental Results
Based on the Weka image classification results, we speculated that the classification errors were caused by the similarities between Asus and MSI gaming series products. The AI-generated database were used to mix the features of both brands. Additionally, the sample size in the Weka database was insufficient, lacking enough training data for products with similar features. To improve classification accuracy the data volume needs to increase for these similar products to enhance the model’s ability to accurately distinguish between different brand features.
3.2.6. Expert Qualitative Analysis
The AI-generated images and the classification results of the Weka system were submitted to the design expert for evaluation, and the manual classification results of the experts were compared with the automatic classification results of the system. Table 4 presents a brief introduction to the experts’ data.
Table 4.
Designer Profile.
Table 5.
Designer A’s Design Style Description for PC parts.
Table 6.
Designer B’s Design Style Description for PC parts.
Table 7.
Designer C’s Design Style Description for OP parts.
Table 8.
Designer D’s Design Style Description for OP parts.
3.3. Expert Qualitative Results Analysis
We used Weka tools and expert evaluations to perform product classification comparisons and explored the effectiveness of machine learning applications in eSports product design. Specifically, we conducted detailed classification assessments for graphics cards, chassis, laptops, mice, and headsets.
- Graphics Cards and Chassis: The inconsistency between expert and Weka classifications in these categories, especially the misclassification of Graphics Card 2 and Chassis 6, highlights the cognitive challenges posed by visual style similarities. This indicates that even experts are confused by design elements with subtle differences.
- Laptops: Weka made a classification error with Laptop 6, and designers A and B also misjudged Laptops 2 and 4, reflecting the risk of misjudgment even with significant brand designs, particularly when product designs are close to or mimic competitors.
- Mice and Headsets: The classification errors for mice and headsets demonstrate the limitations of using visual identification to distinguish brand characteristics in highly similar product categories. Particularly in the headset category, Designer B’s misclassification of Headsets 3 and 4 highlights the subjective differences in brand style interpretation even among experts.
Although the Weka classification model was evaluated as effective in helping novice designers, the qualitative analysis by experts showed that professional experience is indispensable in maintaining brand visual consistency and handling subtle style differences. This emphasizes the importance of combining machine learning tools and expert esthetics in the product development process and the need for strengthening machine learning models in detail recognition and style analysis.
4. Conclusions
We classified the images of eSports hardware products and compared the results with professional designers’ evaluations using the Weka tool. The Weka tool accurately classified products with brand-specific features but struggled with products having similar styles or subtle design details. The potential of machine learning in visual product classification was validated by highlighting the need for real-time feedback in design evaluation. The developed model helps designers better understand and utilize brand-specific design languages and enhance product market competitiveness. This machine learning model can be applied to the automotive, consumer electronics, and fashion industries, supporting innovation and market trend analysis. Further research is necessary for optimization and applicability in different market conditions and more efficient data-driven decision-making and product development.
Author Contributions
Conceptualization, C.-Y.C.; methodology, C.-Y.C.; software, C.-Y.C.; validation, C.-Y.C. and H.-H.W.; formal analysis, C.-Y.C.; investigation, C.-Y.C.; resources, C.-Y.C.; data curation, C.-Y.C.; writing—original draft preparation, C.-Y.C.; writing—review and editing, H.-H.W.; visualization, C.-Y.C.; supervision, H.-H.W.; project administration, H.-H.W.; funding acquisition, H.-H.W. All authors have read and agreed to the published version of the manuscript.
Funding
This research was funded by the National Science and Technology Council (NSTC), Taiwan, R.O.C., under grant number 111-2410-H-027-019-MY2.
Institutional Review Board Statement
Not applicable.
Informed Consent Statement
Not applicable.
Data Availability Statement
Data are contained within the article.
Conflicts of Interest
The authors declare no conflict of interest.
References
- Hsu, C.-L.; Chen, Y.-C.; Yang, T.-N.; Lin, W.-K.; Liu, Y.-H. Does product design matter? Exploring its influences in consumers’ psychological responses and brand loyalty. Inf. Technol. People 2018, 31, 886–907. [Google Scholar] [CrossRef]
- Zhang, Y. Research on key technologies of remote design of mechanical products based on AI. J. Vis. Commun. Image Represent. 2019, 60, 250–257. [Google Scholar] [CrossRef]
- Calò, T. Rethinking Design Workflows with Artificial Intelligence. In Proceedings of the CHItaly 2023: Crossing HCI and AI, Doctoral Consortium, Turin, Italy, 20–22 September 2023; pp. 8–14. [Google Scholar]
- Schmitt, B. Experiential marketing. J. Mark. Manag. 1999, 15, 53–67. [Google Scholar] [CrossRef]
- Kapferer, J.-N. The New Strategic Brand Management: Advanced Insights and Strategic Thinking; Kogan Page Publishers: London, UK, 2012. [Google Scholar]
- Chen, W.; Xie, D.; Zhang, Y.; Pu, S. All you need is a few shifts: Designing efficient convolutional neural networks for image classification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, CA, USA, 15–20 June 2019; pp. 7241–7250. [Google Scholar]
- LeCun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature 2015, 521, 436–444. [Google Scholar] [CrossRef] [PubMed]
- Goodfellow, I.; Pouget-Abadie, J.; Mirza, M.; Xu, B.; Warde-Farley, D.; Ozair, S.; Courville, A.; Bengio, Y. Generative adversarial nets. Commun. ACM 2020, 63, 139–144. [Google Scholar] [CrossRef]
- Wang, H.-H.; Chen, C.-P. A case study on evolution of car styling and brand consistency using deep learning. Symmetry 2020, 12, 2074. [Google Scholar] [CrossRef]
- Maher, M.L.; Weisz, J.D.; Chilton, L.B.; Geyer, W.; Strobelt, H. HAI-GEN 2023: 4th Workshop on Human-AI Co-Creation with Generative Models. In Companion Proceedings of the 28th International Conference on Intelligent User Interfaces; Association for Computing Machinery: New York, NY, USA, 2023; pp. 190–192. [Google Scholar]
- Weisz, J.D.; Maher, M.L.; Strobelt, H.; Chilton, L.B.; Bau, D.; Geyer, W. Hai-gen 2022: 3rd workshop on human-ai co-creation with generative models. In Proceedings of the 27th International Conference on Intelligent User Interfaces, Helsinki, Finland, 22–25 March 2022; pp. 4–6. [Google Scholar]
- Agersborg, C.; Månsson, I.; Roth, E. Brand Management and Artificial Intelligence—A World of Man Plus Machine: A Qualitative Study Exploring How Artificial Intelligence Can Contribute to Brand Management in the B2C Sector. Master’s Thesis, University of Gothenburg, Gothenburg, Sweden, 2020. Available online: https://gupea.ub.gu.se/handle/2077/64924 (accessed on 1 July 2024).
- Derevyanko, N.; Zalevska, O. Comparative analysis of neural networks Midjourney, Stable Diffusion, and DALL-E and ways of their implementation in the educational process of students of design specialities. Sci. Bull. Mukachevo State Univ. Ser. Pedagog. Psychol. 2023, 9, 36–44. [Google Scholar] [CrossRef]
- Wasielewski, A. “Midjourney can’t count”: Questions of representation and meaning for text-to-image generators. Interdiscip. J. Image Sci. 2023, 37, 71–82. [Google Scholar]
- Borji, A. Generated faces in the wild: Quantitative comparison of stable diffusion, midjourney and dall-e 2. arXiv 2022, arXiv:2210.00586. [Google Scholar]
- Caggiani, L.; Camporeale, R. Toward sustainability: Bike-sharing systems design, simulation and management. Sustainability 2021, 13, 7519. [Google Scholar] [CrossRef]
- Wang, C.-H.; Yun, Y. Demand planning and sales forecasting for motherboard manufacturers considering dynamic interactions of computer products. Comput. Ind. Eng. 2020, 149, 106788. [Google Scholar] [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).