Ferromagnetic foreign bodies accidentally inserted in patients usually need to be surgically removed. The methods conventionally employed for locating foreign bodies are often ineffective due to the low accuracy in determining the position of the object and pose risks arising from the exposure of medical staff and patients to ionizing radiation during long-term procedures. New methods using SQUID sensors successfully located foreign bodies in an innocuous and noninvasive way, but they have the drawback of presenting high cost and low portability. This work is part of new research that seeks to bring greater portability and low cost in locating foreign bodies in the human body using GMI and GMR sensors. The main objective of this work is to evaluate and apply the use of Deep Learning in the development of a portable and manual device based on a GMR sensor, including position tracking and orientation of this device from images of known patterns obtained by a camera integrated to the device and the solution of the inverse magnetic problem from the obtained magnetic mapping. The techniques presented are capable of tracking the device with good accuracy and detecting the localization of the foreign body with similar or better results than those obtained in previous works, depending on the parameter. The results obtained are promising as a basis for future developments.