This thesis treats the development of a metal detector. At first the transducer is threaded, together with the response of different materials. A series of tests is made to verify the model and it is reasonable accurate. The next part of the thesis treats the extraction of features from the signals measured on the transducer and a classification based on these features. The features extracted are phase and gain for the transducer. Several methods for this feature extraction are found and thoroughly tested. The classification are made with a classic PCA. The last part of the thesis treats the mapping of the feature extraction and the classification to a FPGA in a hardware/ software co-design process. The mapping are done in to steps, where the first step treats the algorithm level and the second step threats the architecture level. The system is finally tested and while it does not seem to give a significant improvement with the normal benchmark methods, where the tests are made with metal alone. It does give a significant improvement in the ability to detect metal that are mixed with a product.