English title

---

A new automatic target recognition algorithm to recognize and distinguish three classes of targets: personnel, wheeled vehicles and animals, is proposed using a low-resolution ground surveillance pulse Doppler radar. The Chirplet transformation, a time frequency signal processing technique, is implemented in this paper. The parameterized RADAR signal is then analyzed by the Zernike Moments (ZM) in order to feature extraction and they trained to the Nearest Neighbor classifier. The current work provides a new approach for multi-resolution analysis and classification of non stationary signals with the objective of revealing important features in noisy and cluttered environment. The algorithm is trained and tested on real RADAR signatures of multiple examples of moving targets from each class. Results show that the classifier is invariant to target aspect angle and speed.

---

Handwritten digit recognition can be categorized as a classification problem. Probabilistic Neural Network (PNN) is one of the most effective and useful classifiers, which works based on Bayesian rule. In this paper, in order to recognize Persian (Farsi) handwritten digit recognition, a combination of intelligent clustering method and PNN has been utilized. Hoda database, which includes 80000 Persian handwritten digit images, has been used to evaluate our proposed classifier. Obtained results show that PNN is a powerful classifier and excellent choice for classification of Persian handwritten digits. Correct recognition rate when training and testing data have been used directly (without clustering) for training data is 100% and for testing data is 96%, but when k-means has been used as cluster tool and clusters' center have been used as training data, in this case, correct recognition rate for training data is 100% and for testing data is 96.16%. In addition, when Particle Swarm Optimization (PSO) has been used to find optimum clusters for each class of Persian handwritten digits, correct recognition rate in training data is 100% and for the testing data it reaches to 98.18%.

---

In this paper, pattern recognition algorithms are employed to detect and classify the type of high impedance faults (broken and unbroken) and in case of broken ones to determine the surface (gravel, asphalt and concrete) which the conductor has become in contact with it in power distribution networks.

These methods are multilayer SVM and Fuzzy ART classifiers on the bases of features extracted by S-transform and TT-transform from feeder one cycle post-fault current waveforms. These features include energy, standard deviation and median absolute deviation.

The proposed algorithms have been tested on different data set, obtained from field tests and simulated data for events with similar characteristics. The results have shown that the features which are extracted by applying TT-transform contain more information and separability characteristics than those extracted by S-transform and also Fuzzy ART classifier has more accuracy in comparison with multilayer SVM.

---

One of the most preprocessing steps before the classification of hyperspectral images is supervised feature extraction. Because obtaining the training samples is hard and time consuming, the number of available training samples is limited. We propose a supervised feature extraction method in this paper that is efficient in small sample size situation. The proposed method, which is called weighted feature line embedding (WFLE), uses the feature line concepts for production of virtual training samples and then, uses them for estimation of within-class and between-class scatter matrices. The new idea of WFLE is based on more correction on the non-appropriate and abnormal samples through weighting process in estimation of scatter matrices. The WFLE is compared with some popular and state-of-the-art feature extraction methods such as LDA, GDA, NWFE, NPE, LPP and NFLE. The experimental results show the good performance of WFLE in comparison with other methods in small sample size situation.

---

Car license plate detection has been always a challenging task in the context of traffic control and traffic offenses. In this paper, the problem of license plate detection from gray scale images taken in natural conditions is addressed. Our car plate database consists of images with severe imaging conditions such as low quality images, far distanced cameras, and severe weather conditions. The main proposed idea in this paper, is to learn discriminative dictionaries from plate and non-plate exemplars for the classification task. A well-known dictionary learning algorithm is employed for this purpose. However, initial preprocessing techniques for image enhancement are also applied. The achieved results reveal that sparse representation-based techniques in the proposed method can increase the detection accuracy and improve the previously reported results.

---

Rice classification and detection of its quality as a main field in the modern agriculture is attracted many researchers in recent years. This problem is a major issue in the scientific and commercial fields associated with modern agriculture. Different processing techniques in recent years are applied to recognize various types of agricultural products. There are also several color-based and texture-based features to achieve the desired results in this classification procedure. In this paper, the problem of rice categorization and quality detection is considered using sparse non-negative matrix factorization algorithm. This technique includes non-negative matrix factorization method with sparsity constraint to achieve dictionaries that represent the structural content of rice variety. Also, these dictionaries are corrected in such a way to yield the dictionaries with least coherence values to each other. The results of the proposed classifier based on the learned models are compared with the results obtained from the neural network and support vector machine classifiers. Simulation results show that the proposed method based on the combinational features is able to identify the type of rice grain and determine its quality with high accuracy rate.
 

---

Classification of ECG arrhythmia along with medical knowledge can lead to proper decision-making on the patient's condition. Also, classification of arrhythmia types is one of the challenging issues due to the need for detailed analysis of the extracted feature from ECG signal. Therefore, addressing this field using signal processing techniques can be very important. In this paper, various types of morphological features are used to determine the type of ECG arrhythmia. Sparse structured principal component analysis and sparse non-negative matrix factorization algorithms are used to learn the over-complete models based on the characteristics of each data category. Also, the wavelet packet transform coefficients are calculated in different decomposition level to learn over-complete dictionaries. The results of this categorization are compared with the results of the classification based on the neural network, support vector machine another methods presented in this processing field. The simulation results show that the proposed method based on the selected combinational features and learning the over-complete dictionaries can be able to classify the types of ECG arrhythmia precisely.

---

Electricity utility have long sought to identify and reduce energy theft, which represents significant part of non-technical losses. On the other hand, once a fraudulent customer is detected, on-site inspection is necessary for final verification. Since inspecting all customers is expensive, utilities seek to reduce the range of inspection to cases with a higher probability of theft. One way to reduce the scope of inspection is to use artificial intelligence-based methods. An essential challenge here is data imbalance in terms of the ratio of normal to fraudulent customers, which leads to the poor performance of algorithms. In This paper in order to overcome this challenge, assuming that suspicious behavior can be expressed as a mathematical function of normal behavior, in the first stage, the consumption pattern of normal and suspicious customers is categorized using clustering algorithms. Then a deep neural network is trained to model suspicious customers. Next, using trained network, possible theft scenarios for normal costumers are predicted. Finally, a secondary deep neural network is trained to separate the normal and suspicious customers. Assessment of the proposed algorithm for different scenarios on a real data-set with more than 6000 customers and comparison with previous research shows its high performance.

---

Classification of brain tumors using MRI images along with medical knowledge can lead to proper decision-making on the patient's condition. Also, classification of benign or malignant tumors is one of the challenging issues due to the need for detailed analysis of tumor tissue. Therefore, addressing this field using image processing techniques can be very important. In this paper, various types of texture-based and statistical-based features are used to determine the type of brain tumor and different types of features are applied in this classification procedure. Sparse coding and dictionary learning techniques are used to learn the over-complete models based on the characteristics of each data category. The classification process is carried out based on the calculated energy of the sparse coefficients. Also, the results of this categorization are compared with the results of the classification based on the neural network and support vector machine. The simulation results show that the proposed method based on the selected combinational features and learning the over-complete dictionaries can be able to classify the types of brain tumors precisely.

---

In recent years, machine learning approaches play an important role in quality identification of  manufactured products including tiles and ceramics. Deviation of tile dimensions is one the main challenge in ceramic and tile industry. Prediction of this deformation will be beneficial if it can be predicted before producing the tile. In this paper, an automatic system has been proposed to predict the deviation of the ceramic tiles. Besides, a machine learning approach is utilized to identify the most effective parameters that leads to tiles’ defect. In this way, three different classification approaches including logistic regression, random forest, and support vector machine have been studied and the best solution is determined for this purpose. Moreover, several feature sets and forward feature selection method have been employed to select more effective variables on our decision making. The experimental results conducted on real-world dataset show that, random forest approach achieves better performance than others, and the results illustrate that improper temperature parameters has more effect on tile deviation.

---

This paper presents a new method for predicting fault location on distribution networks based on the support vector machine (SVM) technique. The proposed in an online non model based scheme which works based on the real data provided by wide area signals, performs as an intelligent indicator for online estimation of fault locations in distribution systems. In this case, for training intelligent SVM based indicator, a feature selection technique is used to find the best combination of the system phasor variables as input signal to the relay. For this purpose, several stable/unstable scenarios with the potential of oscillating dynamic behaviors are created by time domain transient stability simulation. The main merit of the proposed protection scheme is its ability for predicting instead of detection which can reasonably increase relay speed. The proposed approach is applied on IEEE 33-bus test system and the simulation results show promising performance for the SVM based relay.
 

---

Pistachio, a flowering plant from the Anacardiaceae family, is categorized into various types based on its physical characteristics. Due to its high market value and nutritional importance, accurate identification and packaging based on the pistachio variety are essential for addressing export challenges. Pistachio classification is often performed by electromechanical machines, but these machines lack the necessary precision and can damage the pistachio kernels. Therefore, there is a growing demand for new technologies to improve pistachio classification and separation. In this study, we used a modified version of the MobileNetV3 deep learning model to identify different pistachio varieties. Additionally, by leveraging the Small version of MobileNet, we can efficiently deploy the trained model on smartphones, as it is optimized for computational efficiency. The research was conducted using a dataset of 2148 images representing the Kerman and Sirt pistachio varieties. To increase the number and diversity of images, data augmentation techniques were applied. This helps prevent overfitting and enables the model to generalize better to unseen data. Our modified MobileNetV3 model achieved an accuracy of 99.30% in identifying the two pistachio varieties, outperforming existing classification methods.