classification my own dataset

[zofnat bar]

I am training a new model classification model based on about 239 images which is divided in to 3 categories (labels). my val 42 images.

when i clasiffy an image- the Prediction is:


  ~/Documents/caffe$ ./build/examples/cpp_classification/classification.bin   models/cookie_model/deploy.prototxt   models/cookie_model/cookie_train_iter_100.caffemodel   data/ilsvrc12/imagenet_mean.binaryproto   data/ilsvrc12/synset_words.txt   examples/images/coockie1.jpeg
---------- Prediction for examples/images/coockie1.jpeg ----------
0.3385 - "n2 revaL-7"
0.3338 - "n3 shoko-chips"
0.3278 - "n1 ad-hazot"


different image :
 ~/Documents/caffe$ ./build/examples/cpp_classification/classification.bin   models/cookie_model/deploy.prototxt   models/cookie_model/cookie_train_iter_100.caffemodel   data/ilsvrc12/imagenet_mean.binaryproto   data/ilsvrc12/synset_words.txt   examples/images/coockie2.jpeg
---------- Prediction for examples/images/coockie2.jpeg ----------
0.3385 - "n2 revaL-7"
0.3338 - "n3 shoko-chips"
0.3278 - "n1 ad-hazot"

another different image  of cat  :

~/Documents/caffe$ ./build/examples/cpp_classification/classification.bin   models/cookie_model/deploy.prototxt   models/cookie_model/cookie_train_iter_100.caffemodel   data/ilsvrc12/imagenet_mean.binaryproto   data/ilsvrc12/synset_words.txt   examples/images/cat.jpg
---------- Prediction for examples/images/cat.jpg ----------
0.3385 - "n2 revaL-7"
0.3338 - "n3 shoko-chips"
0.3278 - "n1 ad-hazot"

i try to change the layer but nothing help.

the accuracy about every pictures is the same all the time, what can i do ?
thanks in advance