Deep convolutional neural networks (CNNs) have attained remarkable success in numerous visual recognition tasks. There are two challenges when adopting CNNs in real-world applications: a) Existing CNNs are computationally expensive and memory intensive, impeding their use in edge computing; b) there is no standard methodology for designing the CNN architecture for the intended problem. network pruning/compression has emerged as a research direction to address the first challenge, and it has proven to moderate CNN computational load successfully. For the second challenge, various evolutionary algorithms have been proposed thus far. The algorithm proposed in this paper can be viewed as a solution to both challenges. Instead of using constant predefined criteria to evaluate the filters of CNN layers, the proposed algorithm establishes evaluation criteria in online manner during network training based on the combination of each filter’s profit in its layer and the next layer. In addition, the novel method suggested that it inserts new filters into the CNN layers. The proposed algorithm is not simply a pruning strategy but determines the optimal number of filters. Training on multiple CNN architectures allows us to demonstrate the efficacy of our approach empirically. Compared to current pruning algorithms, our algorithm yields a network with a remarkable prune ratio and accuracy. Despite the relatively high computational cost of an epoch in the proposed algorithm in pruning, altogether it achieves the resultant network faster than other algorithms.

Although, speech recognition systems are widely used and their accuracies are continuously increased, there is a considerable performance gap between their accuracies and human recognition ability. This is partially due to high speaker variations in speech signal. Deep neural networks are among the best tools for acoustic modeling. Recently, using hybrid deep neural network and hidden Markov model (HMM) leads to considerable performance achievement in speech recognition problem because deep networks model complex correlations between features. The main aim of this paper is to achieve a better acoustic modeling by changing the structure of deep convolutional neural network (CNN) in order to adapt speaking variations. In this way, existing models and corresponding inference task have been improved and extended. Here, we propose adaptive windows convolutional neural network (AWCNN) to analyze joint temporal-spectral features variation. AWCNN changes the structure of CNN and estimates the probabilities of HMM states. We propose adaptive windows convolutional neural network in order to make the model more robust against the speech signal variations for both single speaker and among various speakers. This model can better model speech signals. The AWCNN method applies to the speech spectrogram and models time-frequency varieties. This network handles speaker feature variations, speech signal varieties, and variations in phone duration. The obtained results and analysis on FARSDAT and TIMIT datasets show that, for phone recognition task, the proposed structure achieves 1. 2%, 1. 1% absolute error reduction with respect to CNN models respectively, which is a considerable improvement in this problem. Based on the results obtained by the conducted experiments, we conclude that the use of speaker information is very beneficial for recognition accuracy.

Speaker recognition is a process of recognizing persons based on their voice which is widely used in many applications. Although many researches have been performed in this domain, there are some challenges that have not been addressed yet. In this research, Neutrosophic (NS) theory and convolutional neural networks (CNN) are used to improve the accuracy of speaker recognition systems. To do this, at first, the spectrogram of the signal is created from the speech signal and then transferred to the NS domain. In the next step, the alpha correction operator is applied repeatedly until reaching constant entropy in subsequent iterations. Finally, a convolutional neural networks architecture is proposed to classify spectrograms in the NS domain. Two datasets TIMIT and Aurora2 are used to evaluate the effectiveness of the proposed method. The precision of the proposed method on two datasets TIMIT and Aurora2 are 93.79% and 95.24%, respectively, demonstrating that the proposed model outperforms competitive models.

Finger-Knuckle-Print (FKP) is an accurate and reliable biometric in compare to other hand-based biometrics like fingerprint because of the finger's dorsal region is not exposed to surfaces. In this paper, a simple end-to-end method based on convolutional neural network (CNN) is proposed for FKP recognition. The proposed model is composed only of three convolutional layers and two fully connected layers. The number of trainable parameters hereby has significantly reduced. Additionally, a straightforward method is utilized for data augmentation in this paper. The performance of the proposed network is evaluated on Poly-U FKP dataset based on 10-fold cross-validation. The best recognition accuracy, mean accuracy and standard deviation are 99.83%, 99.18%, and 0.76, respectively. Experimental results show that the proposed method outperforms the state-of-the-arts in terms of recognition accuracy and the number of trainable parameters. Also, in compare to four fine-tuned CNN models including AlexNet, VGG16, ResNet34, and GoogleNet, the proposed simple method achieved higher performance in terms of recognition accuracy and the numbers of trainable parameters and training time.

Due to the cursive-ness and high variability of Persian scripts, the segmentation of handwritten words into sub-words is still a challenging task. These issues could be addressed in a holistic approach by sidestepping segmentation at the character level. In this paper, an end-to-end holistic method based on deep convolutional neural network is proposed to recognize off-line Persian handwritten words. The proposed model uses only five convolutional layers and two fully connected layers for classifying word images effectively, which can lead to a substantial reduction in parameters. The effect of various pooling strategies is also investigated in this paper. The primary goal of this article is to ignore handcrafted feature extraction and attain a generalized and stable word recognition system. The presented model is assessed using two famous handwritten Persian word databases called Sadri and IRANSHAHR. The recognition accuracies were obtained at 98.6% and 94.6%, on Sadri and IRANSHAHR datasets respectively, and outperformed the state-of-the-art methods.

One of the active areas of research in concrete structure health monitoring is the detection of cracking in structural elements. Image classification and diagnosis have attracted the attention of many researchers nowadays. Due to the advancement of artificial neural networks and their fast processing, a convolution neural network has been established to detect these cracks. In this study, crack detection in concrete structures has been studied using a convolutional neural network, which can be generalized to all concrete structures for example dams, canals, bridges, shells, road infrastructure, foundations and concrete frames. Convolution neural network training was performed by the SGDM method with the ReLU activator function. Also, 250 iterations were employed for convolution neural network training, which gradually reduced the error rate and increased the accuracy of detecting cracked and uneaten concrete. The convolutional neural network is trained and validated with these 250 iterations. First, images with 32-pixel window dimensions are converted and separated. Then, the 32-pixel window, the 16-pixel, and the 8-pixel windows filter the images. A total of 3 stages of 32, 16, and 8-pixel filter images are analyzed and interpreted. During the training process, validation is performed every 20 iterations, and a diagram related to the accuracy of convolution network estimation and data classification error is drawn and completed. In convolutional neural networks, where the output is in pairs, the cracked and uncracked images of the network architecture are almost identical, differing only in minor specifications. The database of this research includes 20,000 images of cracked concrete and 20,000 uncracked concrete with dimensions of 3×227×227 pixels, 80% of it is used for training and the remaining 20% is used for validation of the convolution neural network. The accuracy of distinguishing cracked concrete from uncracked ones is about 98.16%, which is acceptable for operation and is considered practical. To evaluate the accuracy and performance of the proposed algorithm, each classification was performed against the overall accuracy, the confusion matrix was used for the validation data. According to the clutter matrix, 3861 images, in other words, 48.3% have been predicted to be correctly cracked, and 3992 images, equivalent to 49.9%, have been predicted to be correctly uncracked, and a total of 147 incorrect images have been predicted, which is equivalent to 1.8 percent. Images that are cracked and not accidentally cracked are predicted. They had crack lines in the corner of the image or cracks with a very small width, which the proposed convolutional neural network was mistaken for due to a very small crack width or crack position. Also, the results of the present study showed that the accuracy of this research has the best accuracy in less analysis time compared to previous studies. It should be noted that this method and its associated database can be used to produce a crack detection application on a smartphone, to be able to make a good initial estimate of the structure in question, such as a bridge or building after an unusual loading event, such as an earthquake or explosion.