SQL injection is a hacking attack where malicious code is inserted into database queries through user inputs like search boxes, login forms, or URL parameters. These attacks pose a significant threat to web applications and ERP systems, making early detection crucial. Traditional detection methods, such as rule-based and signature-based approaches, rely on known SQL injection patterns. However, they often fail to identify novel, obfuscated, or zero-day attacks, highlighting the need for more adaptive and intelligent detection mechanisms. This research proposes FusionNetSQL, a fusion-based deep learning model that combines Convolutional Neural Networks, Long Short-Term Memory networks, and Transformers to detect SQL injection attacks. By integrating these architectures, FusionNet-SQL gains a comprehensive understanding of SQL queries, enabling it to differentiate between legitimate interactions and malicious injections. The CNN captures local patterns, the LSTM models sequential dependencies, and the Transformer enhances global context understanding. The model achieves high performance, with 98.02% accuracy, 99.39% precision, 96.79% recall, 98.07% F1-score, and 98.07% AUC-ROC. With its robust performance and adaptability, FusionNet-SQL offers a powerful solution for securing web applications and ERP systems against SQL injection attacks. Its ability to detect both straightforward and sophisticated attacks makes it well-suited for real-world deployment, reinforcing database security and protecting critical data. This research marks a significant step forward in combating evolving cybersecurity threats.