This work presents a novel two-stage technique for categorising and locating bone abnormalities in X-ray images. The suggested approach concentrates on the upper extremities and takes into account the bones in the humerus, forearm, wrist, elbow, shoulder, hand, and finger. To achieve optimal performance with the least amount of computing work, it combines eight original models that were created from scratch and were influenced by the InceptionV3 model. In the initial step, the area of the bone X-ray image is classified into one of seven groups. Following that, seven classifiers were trained to recognise irregularities in bones sent in the image. Eight models are therefore used in the classification step: one for categorisation and seven for abnormality recognition. The study evaluated the efficacy of the proposed method using the largest publicly accessible dataset of bone X-ray scans, the MURA database. The outcomes showed that, while retaining excellent accuracy levels, the suggested approach significantly cuts down on computation and processing time. Furthermore, the hierarchical structure of the system enables the simultaneous examination of bone categorisation and anomaly detection problems. This characteristic sets the suggested method apart from earlier research. Furthermore, this work is the first to incorporate all seven bone sections into the same system, providing a thorough way for classifying and identifying abnormalities in bones in X-ray pictures. Overall, the suggested methodology provides good levels of accuracy and reliability for classifying and identifying abnormalities in bone X-rays.