Yağ Gülü Hasat Durumunun Belirlenmesi için Siyam Sinir Ağları: Yeni Nesil Bir Bitki Tanıma Sistemi

Öz

Bitki biliminde önemli bir alan olan yağ gülünün hasat durumunu belirlemek, doğal yaşamın anlaşılması ve korunmasında kritik bir rol oynar. Geleneksel yöntemlerle yağ gülünün hasat durumunun sınıflandırılması ve tanınması oldukça karmaşıktır. Bu problemi çözmek amacıyla Evrişimsel Sinir Ağı (ESA) tabanlı yaklaşımlar, yağ gülünün hasat durumunu belirlemede başarılı sonuçlar sergilemiştir. Ancak, veri setindeki görüntü sayısının az olması, ESA yaklaşımlarının istenilen performans seviyesine ulaşmalarını engellemektedir. ESA’nın bir türü olan Siyam Sinir Ağları (SSA), bu zorluğa yenilikçi bir çözüm sunmaktadır. SSA, her bir görüntüyü tanımlayan benzersiz özellik vektörlerini çıkartmakta ve daha sonra bu özellik vektörleri bir mesafe ölçütü kullanılarak karşılaştırılmaktadır. Sonuç, benzerlik veya farklılık skoruna göre değerlendirilmektedir. Çalışmanın amacı, SSA ile yağ gülü bitkisinin hasat durumunun belirlenmesidir. Çalışmada modellerin değerlendirilmesinde önceden eğitilmiş VGG16 ve VGG19 modelleriyle birlikte farklı kayıp fonksiyon modelleri ile optimizasyon yöntemlerinin kombinasyonları değerlendirilmiştir. Çalışmada Kosinüs benzerliği mesafe ölçütü olarak kullanılmıştır. Deneyler, herkese açık bir veri seti olan Isparta Gulu (Rosa Damascena Mill.)’nde gerçekleştirilmiştir. Yağ gülü hasat durumu sınıflandırma doğruluğu en yüksek, önerilen SSA-VGG19, Karşılaştırmalı kayıp fonksiyonu ve RMSprop optimizasyon modelindedir. Önerilen bu modelin doğruluk değeri 0,986 ve eğri altında kalan alan (AUC) değeri 0,990 oranlarındadır. Deneyler, yağ gülü hasat durumunun tespitinde önerilen modelin etkili olduğu göstermektedir.

Anahtar Kelimeler

• Yağ gülü
• hasat durumu belirleme
• siyam sinir ağları

Siamese Neural Networks for Determining Oil Rose Harvest Status: A Next-Generation Plant Recognition System

Öz

Determining the harvest status of oil roses, an important area of botany, plays a critical role in understanding and conserving natural life. Traditional methods for classifying and detecting the harvest status of oil roses are quite complex. To solve this problem, Convolutional Neural Network (CNN)-based approaches have shown successful results in determining the harvest status of oil roses. However, the small number of images in the data set prevents CNN approaches from achieving the desired level of performance. Siamese Neural Networks (SNN), a type of CNN, provides an innovative solution to this challenge. SNNs extract unique feature vectors describing each image and then compare these feature vectors using a distance metric. The result is evaluated based on a similarity or difference score. The aim of this study is to use SNNs to determine the harvest status of the oil rose plant. In this study, different combinations of pre-trained VGG16 and VGG19 models with different loss function models and optimization methods were evaluated. Cosine similarity was used as the distance metric. The experiments were conducted on the publicly available Isparta Gulu (Rosa Damascena Mill.) data set. The highest accuracy for the classification of the harvest status of oil roses was achieved with the proposed SSN-VGG19, Contrastive loss function and RMSprop optimization model. The accuracy of the proposed model is 0.986 and the area under the curve (AUC) is 0.990. The experiments indicate that the proposed model is effective in detecting the harvest status of oil roses.

Anahtar Kelimeler

• Oil rose
• harvest status determination
• siamese neural networks

Kaynakça

1. Timor AN. World production oil rose and rose oil. Nature Sciences 2011; 6 (2): 93-110.
2. Duman B, Kayaalp K. Yağ gülü (Rosa damascena Mill.) bitkisinin hasat durumunun makine öğrenmesi ve derin öğrenme yöntemleri ile tespiti. El-Cezeri 2022; 9 (4): 1328-1341.
3. Eli-Chukwu NC. Applications of artificial intelligence in agriculture: A review. Engineering, Technology & Applied Science Research 2019; 9 (4): 4377-4383.
4. Abdelmigid HM, Baz M, AlZain MA, Al-Amri JF, Zaini HG, Abualnaja M, Morsi MM, Alhumaidi A. Spatiotemporal deep learning model for prediction of taif rose phenotyping. Agron J 2022; 12 (4): 807.
5. Thuseethan S, Vigneshwaran P, Charles J, Wimalasooriya C. Siamese network-based lightweight framework for tomato leaf disease recognition. arXiv preprint 2022; arXiv:220911214.
6. Malik M, Ikram A, Batool SN, Aslam W. A performance assessment of rose plant classification using machine learning. Intelligent Technologies and Applications: First International Conference, INTAP 2018; 23-25 October 2018; Bahawalpur, Pakistan, pp. 745-756.
7. Malik M, Aslam W, Nasr EA, Aslam Z, Kadry S. A performance comparison of classification algorithms for rose plants. Comput Intell Neurosci 2022.
8. Sobolu R, Stanca L, Pusta D, Pop I, Cordea M. Image processing technique applying to detect black spot and rust diseases at roses. Managerial Challenges of the Contemporary Society Proceedings 2019; 12 (1): 68-73.

9. Sazzad S, Rajbongshi A, Shakil R, Akter B, Kaiser MS. RoseNet: Rose leave dataset for the development of an automation system to recognize the diseases of rose. Data Brief 2022; 44: 108497.
10. Bhaskar S, Kumar P, Avinash M, Harshini S. Real time farmer assistive flower harvesting agricultural robot. 2021 6th International Conference for Convergence in Technology (I2CT); 2-4 April 2021, Pune, India, pp. 1-8.
11. LeCun Y, Bengio Y, Hinton G. Deep learning. nature, 2015; 521 (7553): 436-444.
12. Fei Y, Li Z, Zhu T, Ni C. A lightweight attention-based convolutional neural networks for fresh-cut flower classification. IEEE 2023; 11: 17283-17293.
13. Saw S, Mahato A, Kumar B, Kukreja V. Rose Multiclassification: Harnessing Hybrid CNN and Random Forest Model. In 2024 International Conference on Automation and Computation (AUTOCOM); 14-16 March 2024, Dehradun, India, pp. 38-41.
14. Mujahid F, Chowdhury PK, Zaman TB, Rahman MM, Reza MT, Nasir NA, Quader MA, Quader MA. Classification of Mixed Color Rose Types Using Convolutional Neural Network. In 2023 5th International Conference on Sustainable Technologies for Industry 5.0 (STI); 09-10 December 2023, Dhaka, Bangladesh, pp. 1-5.
15. Liu C-F, Padhy S, Ramachandran S, Wang VX, Efimov A, Bernal A, Shi L, Vaillant M, Ratnanather JT, Faria AV. Using deep Siamese neural networks for detection of brain asymmetries associated with Alzheimer’s disease and mild cognitive impairment. Magn Reson Imaging 2019; 64: 190-199.
16. Bromley J, Guyon I, LeCun Y, Säckinger E, Shah R. Signature verification using a siamese time delay neural network. Adv Neural Inf Process Syst 1993; Denver, Colorado, USA, pp. 737-744.
17. Shalaby M, Belal NA, Omar Y. Data clustering improves siamese neural networks classification of Parkinson’s disease. Complexity 2021; 2021: 1-9.
18. Yaşar İD, Çakır H, Coşkun A. Siyam sinir ağları ve yerel ikili örüntü kullanılarak temassız avuç içi doğrulaması. J Polytech 2023; 26 (4): 1475-1483.
19. Arısoy MV. Signature verification using siamese neural network one-shot learning. International Journal of Engineering and Innovative Research 2021; 3 (3): 248-260.
20. Toğaçar M, Cömert Z, Ergen B. Siyam sinir ağlarını kullanarak Türk işaret dilindeki rakamların tanımlanması. Dokuz Eylül Üniversitesi Mühendislik Fakültesi Fen ve Mühendislik Dergisi 2021; 23 (68): 349-356.
21. Koch G, Zemel R, Salakhutdinov R. Siamese neural networks for one-shot image recognition. ICML deep learning workshop 2015.
22. Güçlü E, Aaydın İ, Akın E. DCGAN ve Siyam Sinir Ağını Kullanarak Demiryolu Bağlantı Elemanlarındaki Kusurların Tespiti. Demiryolu Mühendisliği 2022; (15): 46-59.
23. Madhu G, Bharadwaj BL, Rohit B, Vardhan KS, Kautish S, Pradeep N. Convolutional Siamese networks for one-shot malaria parasite recognition in microscopic images. In: Demystifying Big Data, Machine Learning, and Deep Learning for Healthcare Analytics Eds, Elsevier, 2021, pp. 277-306.
24. Alakuş TB. Prediction of Monkeypox on the Skin Lesion with the Siamese Deep Learning Model. Balkan Journal of Electrical and Computer Engineering 2023; 11 (3): 225-231.
25. Öztürk Ş. Hash code generation using deep feature selection guided siamese network for content-based medical image retrieval. Gazi U J Sci 2021; 34 (3): 733-746.
26. Lu Y. Research on Small Sample Apple Defect Classification Method Based on Siamese Network. 2023 4th International Conference on Computer Vision, Image and Deep Learning (CVIDL); 12-14 May 2023, Zhuhai, China, pp. 489-493.
27. Duman B. 2022. Erişim Tarihi: 02.02.2024 IspartaGulu(RosadamascenaMill.)_Dataset. https://www.kaggle.com/datasets/1684654f84496eabe23b1728faab7cb9f687086d290f853978658746b65e65e8.
28. Liu C, Miyauchi H, Hayashi K. DeepSniffer: A meta-learning-based chemiresistive odor sensor for recognition and classification of aroma oils. Sens Actuators B 2022; 351: 130960.
29. Lakshmi TS, Govindarajan M, Srinivasulu A. Embedding and Siamese deep neural network-based malware detection in Internet of Things. Int J Pervasive Computing Commun 2022.
30. Meddad M, Moujahdi C, Mikram M, Rziza M. Convolutional siamese neural network for few-shot multi-view face identification. SIViP 2023; 17 (6): 3135-3144.
31. Lachgar M, Hrimech H, Kartit A. Transfer learning for plants’ disease classification with siamese networks in low data regime. International Journal of Computer Engineering and Data Science (IJCEDS) 2021; 1 (1): 8-13.
32. He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. In: Proc IEEE Conf Comput Vis Pattern Recognit 2016; 27-30 June 2016, Las Vegas, NV, USA, pp. 770-778.
33. Alirezazadeh P, Schirrmann M, Stolzenburg F. Improving deep learning-based plant disease classification with attention mechanism. Gesunde Pflanz 2023; 75 (1): 49-59.
34. Vasconcellos ME, Ferreira BG, Leandro JS, Neto BF, Cordeiro FR, Cestari IA, Gutierrez MA, Sobrinho A, Cordeiro TD. Siamese convolutional neural network for heartbeat classification using limited 12-lead ECG datasets. IEEE 2023; 11: 5365-5376.