Impacts of Kernel Size on Different Resized Images in Object Recognition Based on Convolutional Neural Network

This paper focuses a study on impacts of kernel sizes on different resized image relying on convolutional neural network (CNN) for object recognition. Two sets of convolutional neural network (CNN) deep learning models: Conv573 and Conv3, based on shallow feed-forward-architecture network are employed here for feature extraction with comparative assessment purpose. The Conv573 refers to CNN with kernel size of 7×7, 5×5 and 3×3; whilst the Conv3 represents that of only 3×3 kernel size; where three and two convolutional layers of 3×3 kernel size are consecutively comparable to one convolutional layer of 7×7 and 5×5 kernel sizes. The experiments rely on different-resized CIFAR-10 image dataset, 50×50, 100×100 and 150×150 pixels for testing with the Conv573 and Conv3 models. For the purpose of bias reduction, recognition performance assessments depend on averages of precision, recall, F1 and accuracy rates, based upon 10-fold cross validation. The results indicate that the greater the size of an image is, the better the recognition accuracy would be based on Conv573, conversely for Conv3. 2.65% and 0.06% recognition accuracy improvement based on Conv573, whereas 0.39% and 1.76% performance decrease based on Conv3 are respectively yielded when resizing image from 50×50 to 100×100 and from 100×100 to 150×150. For 50×50 and 100×100 resized images, Conv3 yields 4.74% and 1.64% better averaged accuracy than Conv573; nevertheless, Conv573 generates 0.18% better averaged accuracy than Conv3 on 150×150 resized ones.