YAYINLAR

Makine öğrenmesi – Teorik Yönleri ve Python Uygulamaları ile Bir Yapay Zekâ Ekolü

Yapay zekânın bir alt alanı olarak ifade edilen makine öğrenmesi mühendislik, finans ve biyoinformatik’in başı çektiği birçok alanda yaygın olarak kullanılmaktadır. Makine öğrenmesi uygulamaları geliştirmek için temelinde kalkülüs, doğrusal cebir ve istatistik barındıran bazı algoritmaların teorik olarak kavranması önemlidir. Bu algoritmaların teorik yönleri öğrenildikten sonra Python gibi kolay ve zengin kütüphane yapısına sahip bir programlama dili ile kodlanarak uygulama geliştirilebilir. Kitaptaki makine öğrenmesi algoritmalarının teorik yönleri titizlikle irdelenmiş, gerek duyulan doğrusal cebir ve istatistik konuları da özet olarak incelenmiştir. Özgün veri setleri içeren problemler kullanılarak her algoritma için Python uygulamaları geliştirilmiştir. Makine öğrenmesinin bir alt alanı olan Derin Öğrenme ile uygulama geliştirmek isteyen kişilerin de özellikle bu kitaptaki temel bilgileri öğrenmesi önemli bir alt yapı oluşturmalarını sağlayacaktır. Bu kitabı okuduktan sonra derin öğrenme mimarilerinin anlaşılması daha kolay olacaktır.

Kaynak kodlar…

Disease detection and physical disorders classification for citrus fruit images using convolutional neural network

Citrus is an agricultural product with significant added value in the world. Depending on market demands and causing possible economic losses, disease detection in citrus fruits at an early stage is important as well as for all agricultural products. Therefore, it is critical to detect both early-stage citrus diseases and physically damaged citrus fruits with technological solutions. In this study, a new convolutional neural network (CNN)-based model, CitrusNet, was proposed for the classification of defective and deformed citrus fruits. In this study, 5149 images of citrus fruits were collected from citrus orchards in Antalya, Turkey. In the experiments conducted with four different CNN models, the CitrusNet and ResNet50 models obtained the best classification results. In the other phase of the study, experiments were carried out to detect Alternaria alternata and Thrips diseases, which are common in Turkey, using five different CNN models. In the first phase of the study, a dataset of 3582 images of citrus diseases was obtained. The experimental results show that the YOLOv5 and Mask R-CNN models adequately detected the citrus diseases compared to the other models. These models achieved the best performance with an average precision (AP) of 0.99.

UĞUZ S., Gülhan Ş., Abdullah Y. (2023)

A Hybrid CNN-LSTM Model for Traffic Accident Frequency Forecasting During the Tourist Season

Population density in major tourist centers of the world increases significantly during the tourist season. Estimating the frequency of traffic accidents during the upcoming tourist season is of particular interest to many stakeholders, such as local governments. The objective of this study is to propose a hybrid deep learning model, based on convolutional neural network (CNN) and long short term memory (LSTM) models to predict the frequency of traffic accidents during the tourism season. The dataset used in the study includes daily frequencies of traffic accidents with fatalities and injuries that occurred in Antalya between January 2012 and December 2017. In the next phase of the study, seasonal autoregressive integrated moving average (SARIMA), Facebook prophet and deep learning methods including LSTM and the proposed Hybrid CNN-LSTM were tested to predict traffic accident frequencies in Antalya. The experimental results show that the root mean square error (RMSE) of the proposed model is less than 2480, 13266 and 186 compared to SARIMA, prophet and LSTM models, respectively. Also, the R-squared value of the proposed model is greater than 0.016, 0.103 and 0.001 compared to SARIMA, prophet and LSTM models, respectively. It is clear that the proposed hybrid CNN-LSTM model was more successful in predicting traffic accidents when compared to the other models.

UĞUZ S., Erdem B. (2022)

Development of CNN-based GUI for detection of non-motorized vehicles

Today, various solutions are offered for traffic density. One of these suggestions is to popularize the use of bicycles in the category of non-motorized vehicles. For this, first of all, bicycle paths must be built. The use of bicycle lanes or the rate of bicycle use in normal traffic is an important data. Deep learning techniques, which have been popular in recent years, can be used to obtain this data. The aim of this study is to present a model that detects bicycles using various convolutional neural networks architectures. First of all, 962 open source bicycle images obtained from the internet are labeled. For this, trainings were conducted with YOLOv3, YOLOF, Faster R-CNN and Sparse R-CNN architectures. As a result of the trainings, a value of 0.92 mAP was reached with Faster R-CNN. At the end of the study, a software that detects bicycles in real time has been developed.

UĞUZ S., Oğulcan Ç. (2022)

A deep learning based system for real-time detection and sorting of earthworm cocoons

Vermicompost, created by earthworms after eating and digesting organic waste, plays an important role as an organic fertiliser in sustainable agriculture. In this study, a deep learning-based smart system was developed to separate earthworm cocoons used in the production of vermicompost from the compost and return it to production. In the first stage of the study, a dataset containing 1000 images of cocoons was created. The cocoons in each image were labeled and training was performed using a deep learning architecture, one-stage and two-stage models. The models were trained over 2000 epochs with a learning rate of 0.01. From the experimental results, faster R-CNN with ResNet50-FPN model detected the earthworm cocoons better compared to other models. The best performance was obtained by this model with an average precision (AP) of 0.89. In the other stage of the study, the cocoons detected by the software were separated from the compost using a specially designed conveyor belt system. In this process, the detected cocoons are separated from the compost using 10 pneumatic valves that spray air at the separation point. The study is the first of its kind that enables earthworm cocoons to be returned to production with the use of a real-time intelligent system. It also contributes to the literature on small object detection using deep learning.

ÇELİK A., UĞUZ S. (2022)

Classification of olive leaf diseases using deep convolutional neural networks

In recent years, there have been significant achievements in object classification with various techniques using several deep learning architectures. These architectures are now also used for classification and detection of many plant diseases. Olives are important plant species which are grown in certain regions of the world. The disease types that affect the olive plants vary on the region where it is grown. This study presents a data set consisting of 3400 olive leaves samples which also includes healthy leaves so as to detect Aculus olearius and Olive peacock spot diseases, which are common olive plant diseases in Turkey. This experimental study used transfer learning methods on VGG16 and VGG19 architectures, as well as on our proposed CNN architecture. Effects of data augmentation on performance were one aim of this research. In the experimental studies which applied data augmentation the highest success value in trained models was 95%, whereas in the experiments without data augmentation the highest value was 88%. Another subject of this research is the Adam, AdaGrad, Stochastic gradient descent and RMS Prop optimization algorithms’ effect on the network’s performance. As a result of the conducted experiments, Adam and SGD optimization algorithms were generally observed to generate superior results.

UĞUZ S., UYSAL N. (2021)

Automatic Olive Peacock Spot Disease Recognition System Development by Using Single Shot Detector

Among the artificial intelligence based studies conducted in the field of agriculture, disease recognition methods founded on deep learning are observed to become widespread. Due to the diversity and regional specificity of many plant species, studies performed in this field are not at the desired level. Olive peacock spot disease of the olive plant which grows only in certain regions in the world is a widely encountered disease particularly in Turkey. The aim of this research is to develop an olive peacock spot disease detection system using a Single Shot Detector (SSD) which is one the popular deep learning architectures to support olive farmers. This study presents a data set consisting of 1460 olive leaves samples for the detection of olive peacock spot disease. All of the images of the olive leaves which produced under controlled conditions were collected from Aegean region of Turkey during spring and summer. The data set was trained with different intersection over union (IoU) threshold values using SSD architecture. A 96% average precision (AP) value was obtained with IoU=0.5. As IOU value goes up from 0.5, erroneously classified olive peacock spot disease symptoms growed larger as well. The AP curve becomes flat when between 0.1 and 0.5, and it decreases when greater than 0.5. This analysis showed that the IoU significantly influenced the performance of SSD based model in detection of olive peacock spot disease. In addition to, trainings were performed by employing Pytorch library and a GUI was developed for the SSD based application using PyQt5 which is one of Pyhton’s libraries. Results showed that the SSD was a robust tool for recognizing the olive peacock spot disease.

UĞUZ S. (2020)

İmalat Mühendisliği Eğitimi için Sanal Gerçeklik Sistemi Tasarımı ve Geliştirilmesi

Mühendislik eğitiminin önemli bir parçasını oluşturan laboratuvarlar, birçok avantajın yanı sıra yüksek ilk kurulum maliyetleri, gelişen teknolojiye göre güncellenme gereksinimi ve sınırlı kullanım süreleri gibi bazı dezavantajlara da sahiptir. Son yıllarda gerçek laboratuvarların bir alternatifi olarak düşünülen sanal laboratuvar uygulamaları üzerine önemli çalışmalar yürütülmektedir. Sanal laboratuvar çalışmalarının önemli bir bölümü de sanal gerçeklik teknolojisine dayanmaktadır. Makine mühendisliği ve imalat mühendisliği gibi bölümlerde okutulan bilgisayar destekli imalat derslerinde, computer numerical control (CNC) makinelerini kontrol etmek için uygulama yapma gereksinimi duyulur. Bu çalışmada, bu gereksinimin karşılanması amacıyla sanal gerçeklik teknolojisi kullanarak geliştirilen bir sistem önerilmiştir. Öncelikle, sanal gerçeklik başlığı ve kumanda kolundan oluşan bir sanal gerçeklik donanım seti tasarlanmıştır. Daha sonra ise Unity tabanlı bir simülasyon yazılımı geliştirilmiştir. Bu yazılım ile CNC makinesinde farklı G kodu uygulamalarının sonuçlarının öğrenciler tarafından görülmesi sağlanmıştır. İçerik geliştirilmesinin yanı sıra düşük maliyetli sanal gerçeklik donanımının da üretilmesi, bu çalışmayı benzerlerinden ayıran unsurlardan birisidir.

UĞUZ S., Zehir B. (2020)