Sports Image Classification and Batch Management Using Convolutional Neural Network (CNN)

Abstract

The rapid growth of digital sports media creates a practical requirement for automatically classifying and organizing large image collections. This paper presents the Sports Image Classification and Batch Management system documented in the project report. The desktop application combines a custom Keras convolutional neural network with a MobileNetV2 fallback, OpenCV-based preprocessing, a Tkinter graphical interface, SQLite authentication, and a multithreaded batch organizer. The custom model operates on 150×150 RGB inputs for Cricket, Football, and Basketball, whereas the fallback path uses 224×224 inputs and maps predictions to Cricket, Football, Basketball, Volleyball, and Tennis. Single-image prediction exposes RGB conversion, resizing, normalization, tensor expansion, class probability, and confidence output. Batch mode scans mixed folders, applies a configurable confidence threshold, creates category-specific directories, copies or moves files, routes uncertain samples to an Unclassified folder, and reports progress through a live log and classification-ratio chart. The documented application output classifies a Football image with 95.5% confidence and confirms successful end-to-end inference. The project demonstrates a compact and scalable workflow for sports academies, photographers, media archives, educational institutions, and event-management teams while retaining responsiveness through worker threads and thread-safe Tkinter updates.