Numerous potentials are presented by the Internet of Things, but there are a number of drawbacks as well. IoT devices have recently been more frequently the subject of malware assaults. Deep Learning is a popular technique that is used to identify and classify viruses. Researchers are working to strengthen the security of gadgets that are connected to the Internet in this respect. This approach used the behaviour of malware during run-time in the context of system calls to identify it. The real-time IoT malware samples were given by IOTPOT, a honeypot that replicates a variety of IoT device CPU architectures. From the malicious system calls that are generated, a deep learning algorithm extracts the necessary characteristics. To better understand malware activity, RGB photos were transformed and behavioural data was used to depict the samples. The retrieved system calls were divided into two groups—normal and malicious sequences—using VGG-19 (Visual Geometry Group – 19). The two classes were then assigned to each of the 15 subclasses of malware. The model is made lightweight and computationally efficient utilising a two-step feature extraction method that uses complete vector features for classification and lightweight dynamic features for weighting. The efficiency of deep learning is assessed using a range of performance criteria. In comparison to previously developed approaches, we were able to achieve an average classification accuracy of 97.75%, an increase of 3.7%.