Arbitrary Kernel Embossed Linear Edge Filtered Residual Network Based Sports Ball Classification

Sports ball categorization serves a number of vital purposes across several kinds of business sectors that includes sports analytics, education, recreational activities, and inventory management. These objectives are successfully met by automating and improving categorization processes using cutting-edge technologies like deep learning and computer vision. This promotes creativity and efficiency in a variety of applications. With this motivation, this paper proposes Kernel Embossed Edge Filtered ResNet (KEF-ResNet) that categorizes six classes of sports ball type more effectively. The Sports Ball Dataset that includes 2400 ball images that was used for implementation. Initially, the KEF-ResNet model divides ball images into six groups: football, rugby ball, baseball, cricket ball, bowling ball, and shuttlecock. During preprocessing, image labeling is done to the segregated images. Following the augmentation of data on the labelled ball images, ended with 50,400 ball images. To create Kernel Embossed Edge filtered (KEF) ball images, augmented ball images were subjected to the Kernel Embossed Edge with linear filter technique. The KEF ball images are used with both the proposed KEF-ResNet model and the current CNN techniques to assess the performance. With a high accuracy of 99.7%, the implementation shows that the suggested KEF-ResNet model performs well in the sports ball type classification task.

I. Introduction

Sports ball categorization is necessary in a number of areas, including improving athletic performance and training, streamlining inventory control, and developing captivating user interfaces. It is possible to design a reliable sports ball classification system that makes use of both conventional computer vision techniques and cutting-edge deep learning methods by fusing image processing with machine learning methods [1]. It might be difficult to distinguish between different sports balls since they sometimes have similar colors, patterns, or forms. Variations in backdrops, lighting, and angles may all have an impact on how well categorization algorithms work, which can result in errors. Sports balls that are accurately categorized aid in the analysis of gaming. For example, identifying the differences between a basketball and a soccer ball might reveal information on player usage and performance [2]. The ability to classify various ball types facilitates the gathering of precise data, such as shot precision, control over the ball, and play tactics, which may be utilized to enhance player efficiency and enhance training methods [3]. Sports ball categorization software helps simplify inventory management in sports shops and equipment. It facilitates the effective management of reordering procedures, product type identification, and stock level tracking. Precise ball categorization improves user experience in interactive sports games and simulations by guaranteeing authentic connections with the simulated environment [4]. The remaining portion of this work is structured as follows: A thorough analysis of relevant studies on the deep learning- based categorization of sports ball type is given in Section 2. The proposed KEF-ResNet model is covered in Section 3. The experiment's findings and the performance analysis are presented in Section 4. Section 5 provides a final conclusion and recommendations for further study.