Aedes Aegypti Larvae Detection System Based On Convolution Neural Network Via Transfer Learning
Aedes aegypti mosquitoes are small slender fly species spreading the arbovirus from flavivirus vector through the feeding of the mammals’ blood. The early detection of this species is very important. Once this species turns into adult mosquitoes, the population control becomes more complicated. The...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://eprints.utem.edu.my/id/eprint/24715/1/Aedes%20Aegypti%20Larvae%20Detection%20System%20Based%20On%20Convolution%20Neural%20Network%20Via%20Transfer%20Learning.pdf http://eprints.utem.edu.my/id/eprint/24715/2/Aedes%20Aegypti%20Larvae%20Detection%20System%20Based%20On%20Convolution%20Neural%20Network%20Via%20Transfer%20Learning.pdf http://eprints.utem.edu.my/id/eprint/24715/ https://plh.utem.edu.my/cgi-bin/koha/opac-detail.pl?biblionumber=116893 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Aedes aegypti mosquitoes are small slender fly species spreading the arbovirus from flavivirus vector through the feeding of the mammals’ blood. The early detection of this species is very important. Once this species turns into adult mosquitoes, the population control becomes more complicated. The situation even worse when difficult access places like a water storage tank became one of the favourite breeding places for the Aedes aegypti mosquitoes. Therefore, a technological method is required to assist the operator in the field during the routine inspection of the Aedes aegypti larvae, especially at difficult access places as stated in the report of the World Health Organization (WHO). This research proposed a development of the Aedes aegypti larvae detection system based on the convolutional neural network via the transfer learning method. In this study, a database is created since there is no Aedes aegypti database available online. The database is developed by collecting the Aedes aegypti larvae images in in the same environment of water storage tank. 507 images are set for training dataset, 10 images for validation dataset and 30 images for test dataset. Two different convolutional architectures have been trained in this study, which are Faster-Region Convolutional Neural Network (Faster-RCNN) and Single Shot Multibox Detector (SSD) that applying same region proposal techniques and base network of Inception-v2. Besides, the pre-trained model of the Common Object in the Context dataset has been applied in this training, where the hyper-parameter fine-tune configuration has been implemented in this study. The performance of the generated inference graphs is analysed based on three main aspects, which are the performance during training, validation and test. In order to estimate the generalization gap in the training phase, the cross-entropy loss of the training and the validation for both architectures are obtained so that the optimum capacity can be retrieved from the learning. Meanwhile, in the validation phase, the tracking-based metrics and the perimeter intrusion detection metrics are conducted for several specific learning steps in the validation dataset. The precision-recall curve (PR Curve) also has been implemented in the validation phase, where the curve at the right top angle is proposed as the best model in this study. In the test phase, the test dataset is tested with standard detection metrics. From the results obtained in the training, validation and test analyses, it is observed that the best architecture for the detection of the Aedes aegypti larvae is the Faster-RCNN. The results also indicated that the accuracy of the test results for the Faster-RCNN is 0.9213, while the SSD is 0.6966. Therefore, it can be concluded that the Faster-RCNN is the best model in the detection of the Aedes aegypti larvae. The impact of this study is the proposal of a new method with respect to vision technology, specifically for the Aedes Aegypti larvae prevention and outbreak as highlighted by WHO and sustainable development programme by United Nation. |
|---|