Due to the enormous growth of the Internet, cyberspace faces ever-increasing threats. Therefore, the role of a security framework to handle malware threats is essential in the current era. Some malware files are capable of forming a network of infected systems that are exploited by the attackers to perform many cyber attacks like distributed denial of service (DDOS), Phishing, etc. Malware is also employed to steal critical information from the host systems. Apart from the traditional approaches like static and dynamic analysis that suffer from various challenges such as sensitivity to obfuscation methods and computational overhead, image-based malware detection approaches are also studied by many researchers. The existing approaches need a fixed size image input. Instead, a multi-scale learning method can be utilized to enhance the performance of the detection model. Hence, in this work, spatial pyramid pooling (SPP) based malware variant detection models are proposed and their performance is compared with the existing relevant works. The experiments reveal that the proposed technique produces an accuracy of 99% and it outperforms the existing relevant works.