I have generated segments of an image. Each segment represent one object. How to train these segments to recognize objects?

An image contains multiple objects. Segment is generated for each object. i.e. segmented image contains an object with black bakround which represents mask for other region than object of interest. I want to train Deep learning network for these segments. I have replaced 23rd layer of alexnet with newly created fullyconnectedlayer. when I trained and tested some set of images , test images are classified in classes to which alexnet is already trained. Neither of image is classified in class for which i have trained my images. How to train segmented for objects of given class. how to minimize impact of other classes of alexnet?

[MUHAMMED IRFAN]

Hello,

No one can answer your question without actually seeing what you have done in your code. Please share your code.