دانلود مقاله مدل یادگیری عمیق ترکیبی از شبکه عصبی کانولوشن و مکرر برای تشخیص آتش سوزی جنگل
چکیده
1. مقدمه
2 بررسی مطالعات
3 جزئیات مجموعه داده و پروتکل بررسی
4 روش پیشنهادی
5 نتایج تجربی و تحلیل
6 نتیجه گیری و مطالعات آتی
منابع
Abstract
1 Introduction
2 Literature survey
3 Dataset details and investigation protocol
4 Proposed method
5 Experimental results and analysis
6 Conclusion and future work
References
چکیده
آتش‌سوزی جنگل‌ها تهدیدی جدی برای حیات وحش، محیط‌زیست و تمام بشریت است. این تهدید توسعه سیستم‌های مختلف هوشمند و مبتنی بر بینایی رایانه‌ای را برای تشخیص آتش‌سوزی جنگل‌ها ایجاد کرده است. این مقاله یک مدل یادگیری عمیق ترکیبی جدید برای تشخیص آتش‌سوزی جنگل پیشنهاد می‌کند. این مدل از ترکیبی از شبکه عصبی کانولوشن (CNN) و شبکه عصبی بازگشتی (RNN) برای استخراج ویژگی و دو لایه کاملاً متصل برای طبقه‌بندی نهایی استفاده می‌کند. نقشه ویژگی نهایی به دست آمده از CNN مسطح شده و سپس به عنوان ورودی به RNN تغذیه شده است. CNN ویژگی های مختلف سطوح پایین و همچنین سطح بالا را استخراج می کند، در حالی که RNN ویژگی های مختلف وابسته و متوالی را استخراج می کند. استفاده از CNN و RNN برای استخراج ویژگی در این مقاله برای اولین بار در ادبیات تشخیص آتش سوزی جنگل پیشنهاد شده است. عملکرد سیستم پیشنهادی بر روی دو مجموعه داده آتش در دسترس عموم - مجموعه داده آزمایشگاهی Mivia و مجموعه داده آتش Kaggle ارزیابی شده است. نتایج تجربی نشان می‌دهد که مدل پیشنهادی می‌تواند به دقت طبقه‌بندی بسیار بالایی دست یابد و از نتایج پیشرفته موجود در این زمینه بهتر عمل کند.
توجه! این متن ترجمه ماشینی بوده و توسط مترجمین ای ترجمه، ترجمه نشده است.
Abstract
Forest fire poses a serious threat to wildlife, environment, and all mankind. This threat has prompted the development of various intelligent and computer vision based systems to detect forest fire. This article proposes a novel hybrid deep learning model to detect forest fire. This model uses a combination of convolutional neural network (CNN) and recurrent neural network (RNN) for feature extraction and two fully connected layers for final classification. The final feature map obtained from the CNN has been flattened and then fed as an input to the RNN. CNN extracts various low level as well as high level features, whereas RNN extracts various dependent and sequential features. The use of both CNN and RNN for feature extraction is proposed in this article for the first time in the literature of forest fire detection. The performance of the proposed system has been evaluated on two publicly available fire datasets—Mivia lab dataset and Kaggle fire dataset. Experimental results demonstrate that the proposed model is able to achieve very high classification accuracy and outperforms the existing state-of-the-art results in this regard.
Introduction
Forest fire can potentially result in a large number of environmental disasters, causing vast economical and ecological losses apart from jeopardising human lives. These fires pose a serious threat to people, wildlife, and the environment. To preserve the natural resources and protect the properties and human lives, forest fire detection has become very crucial. It has lead to increasing number of research explorations in this area around the world. Early and accurate detection of forest fires is essential for mitigating the effect of the fire as once a forest fire spreads to a large area it becomes very difficult to control it and might result in a catastrophe. In its early stage any forest fire is relatively small and easy to control. Fire and smoke detector sensors can easily be installed in indoor environments, it is generally not the case for forest areas. Sensors also require the fire to burn for a while before they can be detected. On the contrary, vision based devices can be used to detect fire in real life and they can be deployed in any area by using different means. These systems are also cheap and easy to install.
Conclusion and future work
This article proposes a combination of CNN and RNN based deep learning method for forest fire detection. The evaluation of the performance of the present system has been done on two different public datasets. The proposed forest fire detection system outperforms the existing studies in this regard. The present work also overcomes various drawbacks of the existing systems. It is evident from the high classification accuracy of the present system that the present system can be employed to detect forest fires in the real world scenarios. The present work shall provide fresh insight to the researchers in carrying out the new researches on fire detection using computer vision based techniques.
In future, the attempt will be made to carry out the research work in this problem area by employing other sophisticated deep learning techniques. The plan is also there to develop a fire detection system in non-forest areas, especially fires that occur in residential areas and industrial areas. Other possible future directions of this research work include the exploration of the possibility of employing the proposed model for low resolution satellite images covering large geographical areas.