This paper shows how modified VGG16 and a customized model can be used for a weapon classifier system. The basic concept is to detect weapon type by its shape, length, angle, rotation degree, etc. The layered structure of VGG16 has been primarily used along with some extra layers for step-by-step classification. The images have been preprocessed with image augmentation using parameters such as shear, size, and rotation. The VGG16 model has been used for its prediction accuracy with less time complexity. The VGG19 model has better accuracy compared to the VGG16 but the time complexity is higher. The better accuracy of VGG19 is insignificant and negligible. VGG16 has 1 × 1 convolutional layers. VGG16 has small-sized convolution filters allowing the model to have a large number of weight layers. Instead of using a large number of hyper-parameters, the model uses a 3 × 3 convolution layer and a 2 × 2 max pool layer making the layer arrangement more consistent throughout the architecture. The trained mode acquired a test accuracy of 96.5% when tested with new input images.