DETECTION OF DIFFERENT TYPES OF SKIN DISEASES USING RASPBERRY PI

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 06 | Jun 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1403
DETECTION OF DIFFERENT TYPES OF SKIN DISEASES USING RASPBERRY PI
J.S.Leena jasmine.[1], Rutheeswaran.k[2], Sarath. H[3], Mohan Kumar. K [4], Praveen Kumar.M.S[5]
Professor, Department of ECE, Velammal Engineering College, Chennai, Tamil Nadu [1]
UG Scholars[2,3,4] , Department of ECE, Velammal Engineering College, Chennai, Tamil Nadu
-------------------------------------------------------------------------***----------------------------------------------------------------------
Abstract- To diagnosing the different types of skin
diseases is an exigent process even for better-experienced
skin doctors, while there are several reasons for an increase
in such conflicts, foremost among them is the reduction of
these types of diseases. This paper proposed Deep learning.
Nowadays diagnosing diseases through modern technology
becomes easy to access and convenient. Due to the
emergence of smartphone analysis and providing results in
less time. This system will utilize computational techniques
to analyze, process, and relegate the image data predicated
on various features of the images. Skin images are filtered to
remove unwanted noise and process it for enhancement of
the image. Using a machine-learning algorithm
(Convolutional Neural network) can predict the type of
disease and show in the output to the IOT page of the
predicted disease.
KEYWORDS: IoT Protocol, Skin diseases classification
images, Wi-Fi Image Processing, Noise Removal in the given
image
1. INTRODUCTION:
Medical Image Processing is used widely in the diagnosis of
various diseases. It can use to identify various types of skin
diseases. In the current work transfer, learning was used to
identify the skin diseases melanoma, vitiligo, and vascular
tumors. Melanoma is a skin cancer that occurs from the
pigment-producing cells, melanocytes [7]. Most melanomas
are asymmetrical with an uneven border. The color of
melanoma may have a different shade of brown. A small
pencil eraser-shaped lesion is the warning sign for
melanoma.[4] It is not easy for patients to recognize
melanoma. Melanoma is a serious type of skin cancer.
Diagnosed timely and treated suitably, it can very often cure
with comparatively small surgery alone [8]. An avascular
tumor is a tumor that occurs at the vascular origin. It is a
growth of soft tissue that can be either begin or malignant. It
developed from blood vessels or lymph vessels. It is a
massive and complex type of lesion, especially for doctors
with no or little experience in this field. Vascular tumors can
happen in any part of the body and can be of type being (not
cancer) or malignant (cancer). They may form on theskin,in
the tissues below the skin, and/or in an organ [10]. They
may be found in the tissues below the skin or on the skin.
Vitiligo is a disease in which the skin loses its color because
of the loss of its pigment cells (melanocytes). Melanocytes
are responsible for the color of our skin. Lossofmelanocytes
may result in discolored patches on the skin. The discolored
patches usually increase with time [9]. Thediseasecanaffect
the skin on any part of the body. This disease is non-fatal but
it will affect the patient’s mental state. A Convolutional
Neural network or CNN is a very popular Machine-learning
algorithm in the area of computer vision. Training time and
the amount of data required for CNN are more than
traditional machine learning algorithms. People develop
CNN with great performance and share the details to use in
the future. Interesting use of a convolutional neural network
is that this pre-trained network can be reused for a related
task in another new task [5]. Humans can transfer
knowledge learned from one task. The new task will not be
learned from scratch instead previouslyacquiredknowledge
would use to learn a new task. Transfer learning is a very
popular approach in recent days [9]. It is a process of using
gained knowledge of one network on other related
problems. It is used in a situation where yourdatasethastoo
little data to train a model from first. The process of learning
knowledge and applying it to a similaranddifferentproblem
is called Transfer learning. There are two steps to be
followed Pre-training, Fine Tuning. Pre-Training: In this
process, the training of the network will do with a large data
set [1]-[3]. All the parameters of the network are trained
and, in this way, the model is learned. Fine Tuning: The new
dataset can give to fine-tune the pre-trained CNN. The new
data set may be similar or different from the original data
set. If the new data set is similar and the size of the data set
is, less then only the final layer of the model will train to
keep all other layers fixed to avoid overfitting. If the size of
the data set is big then the whole model may be retrained
with the initial weights of the pre-trained model. If the new
data set is not the same as the old data set, then if the size of
the data set is less the initial layer can be fixed and the
remaining layer can be retrained. In the case where the size
of the data set is big, the whole network can be retrained
with the initial weight of the pre-trained network [6]. The
Internet of things (IoT) is an everlastingly growing network
spanning a gathering of devices, sensors,actuators,andeven
smartphones connected to the internet. These devicessense
and assemble data for combination and analysis for better
decisions applications. The system of the Internet of Things
(IoT) contains an extensive diversity of heterogeneous
devices, reaching from wireless sensors to smart home uses
connected to the global IP network, and is predictable to
include many other types of technologies that are classically
not connected so far. The interactive devicesinIoT,forcedin
terms of memory and power, must agree on different

features of the exchanged data for real transmission
depending on specific protocols
2. EXISTING SYSTEM:
Skin Cancer is the most common cause of death, including
colon cancer, lung cancer, breast cancer, cervical cancer,etc.
skin cancer is one cancer that increases every year.
Considering and analyzing features of cancer images, it
includes asymmetry, border irregularity, compact index,
fractal dimension, edge abruptness, color variation, and
changes in the diameter, is a populartechniqueforanalyzing
patients with skin cancer. To extract and analyze such
features, image segmentation plays an important role in
automatic skin cancer detection systems. In this process,we
propose the image segmentation scheme basedona support
vector machine (SVM) and snake activecontour.SVMisused
to help find the appropriate parameter for the snake
algorithm.
3. PROPOSED SYSTEM:
3.1. Dataset collection:
The lack of proper dermatoscopic data in terms of size and
diversity has been a setback for training neural networks.
The dataset used here is the dataset released by the Harvard
Data verse. The dataset consists of 10015 dermatoscopic
images with a 600x450-pixel resolution. It is a collectionof7
distinct classes. Actinic Keratosis (AKIEC) is beign but
potentially malignant and can turn into a malignant lesion.
Basal cell carcinoma (BCC) and melanoma (MEL) are
malignant. Benign Keratosis (BKL), Dermatofibroma (DF),
Vascular Skin Lesion (VASC), and Melanocytic Nevi(NV)are
benign. Fig. 1 depicts the images of these classes. Half of the
dataset was confirmed by histopathology. Few other
techniques to confirm ground truth were reflectance
confocal microscopy, follow-up, or expert consensus. On
examination of the metadata, ithasbeenfoundthatthere are
only 7470 distinct skin lesions out of the 10015 images. NV,
the largest class, has 6705 images while DF, thesmallest,has
only 115 images.
Fig no 1: PROPOSED SYSTEM
3.2. Data preprocessing:
After an initial exploratory data analysis, duplicate images
have been identified. Aftertheremoval ofduplicateimages,a
validation set has made with a split ratio of 83:17. After this
process, the training set has 9077 images and the validation
set has 938 images. Later data augmentation in the form of
flipping, cropping, and rotating have been applied with the
final target image size set to 224x224 pixels. Stratified
sampling has been used to maintain theinter-classratio.The
training set images later passed through a unique pre-
process function concerning the model architecture used.
Each architecture requires input to the model in a certain
format and the preprocess function helps us transform our
data into such required format.
3.3. Transfer learning technique and training:
Transfer learning has been used as a technique to train
neural networks as they are data-hungry and we only have
limited data despite having additional data through data
augmentation. After taking this architectureasa basemodel,
the top layer has been removed and a Global averagepooling
layer, dropout, and dense layers have been added. Different
thresholds of dropout have been tried in this study. The
training has been implemented on Google Collab using an
NVIDIA Tesla T4 provided. Four different architectures
namely ResNet50, InceptionV3, Exception, and Mobile Net
have been used. Class weights have beencomputedbased on
the individual class frequency to tackle class imbalance.
While training, the weights of the base layers have frozen.
We trained each neural network on 30 epochs and
categorical accuracy has used as a measure for validation.
3.4. Evaluation metrics:
Several evaluation metrics could use for a classification task
but here we take into consideration accuracy, precision,and
recall. The recall is the most important measure here it
represents True Positives and True Negatives.Similarly, and
represent False Positives and False Negatives respectively.
Fig no 2: Raspberry pi connected with Pi Camera
IOT
PAGE
POWER
SUPPLY
SD-CARD
RASPBERRY
PI
PI CAMERA

4. RESULTS AND DISCUSSION:
At first, we have to Open the VNC viewer
Connect the raspberry PI with the desktop Open the file
manager in the VNC viewer in the file manager, we have to
open the RPI vision And in the RPI vision folder, we have to
open the skin diseases The code will run and we have to
execute the code and the last step we havetopress entertwo
times The PI Camera will be on, the diseaseswill beshownto
the pi camera, and the diseases will bedetectedwhichwould
show on the desktop.
CLASSIFICATION RESULTS:
The image classifier that has built in the VNC Viewer has a
Precision of 100%, a Recall of 100%, and an Average
Precision (AP) of 100%. The result that will send will be the
output of the IoT Page
(a) (b)
Fig no 3: Skin diseases images name: a) wheal b) Skin
cancer
IN IoT PAGE Fig no 4: Psoriasis skin diseases:
Fig no 5: Wheal skin diseases:
Fig no 6: IoT page output:
5. CONCLUSION:
Skin disease is a common and frightening kind of disease in
the whole world. The new development of deep learning in
medical imaging gives us the motivation to diagnose skin
disease from digital images. Transfer learning is a very
popular and emerging technology in the field of medical
image processing. We developed a model to classify three
skin diseases. Transfer learning is a very popular and
emerging technology in the field of medical image
processing. We developed a model to classify three skin
diseases melanoma, vitiligo, and vascular tumor. To make
the dataset, many data pre-processing and data
augmentation methods were applied. The proposed model
used the inception V3 model as a base model thatgavea very
good accuracy for training
7. REFERENCES:
1. A Mutual Bootstrapping Model for Automated Skin
Lesion Segmentation and Classification Yutong
Xie;Jianpeng Zhang;Yong Xia;Chunhua ShenYear:
2020
2. Self-Paced Balance Learning for Clinical Skin Disease
Recognition Jufeng Yang;Xiaoping Wu;Jie
Liang;Xiaoxiao Sun;Ming-Ming Cheng;Paul L.
Rosin;Liang Wang Year: 2020
3. H. Gupta, H. Bhatia, D. Giri, R. Saxena, and R. Singh,
Comparison and Analysis of Skin Lesion on
Pretrained Architectures ComparisonandAnalysisof
Skin Lesion on PretrainedArchitecturesno. July2020,
doi: 10.13140/RG.2.2.32161.43367.
4. Discriminative Feature Learning for Skin Disease
Classification Using Deep Convolutional Neural
Network Belal Ahmad ; Mohd Usama; Chuen-Min
Huang; Kai Hwang; M. Shamim Hossain; Ghulam
Muhammad Year: 2020
5. Deep Learning in Skin Disease Image Recognition: A
Review Ling-Fang Li;Xu Wang;Wei-Jian Hu;Neal N.
Xiong;Yong-Xing Du;Bao-Shan Li Year: 2020

6. Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe,
Jonathon Shlens, Zbigniew Wojna, “Rethinking the
Inception Architecture for Computer Vision”.
arXiv:1512.00567
7. François Chollet, "Xception: Deep Learning with
Depthwise Separable Convolutions”
arXiv:1610.02357
8. Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai,
Joan Puigcerver, Jessica Yung, Sylvain Gelly, Neil
Houlsby, “Big Transfer (BiT): General Visual
Representation Learning” arXiv:1912.11370
9. H. B. Muresan, Skin Lesion Diagnosis Using Deep
Learning, Proc. -2019 IEEE 15th Int. Conf. Intell.
Comput. Commun. Process. ICCP 2019, pp. 499506,
2019, doi: 10.1109/ICCP48234.2019.8959661
10. E. H. Mohamed and W. H. El-Behaidy, Enhanced Skin
Lesions Classification Using Deep Convolutional
Networks, Proc. - 2019

DETECTION OF DIFFERENT TYPES OF SKIN DISEASES USING RASPBERRY PI

More Related Content

Similar to DETECTION OF DIFFERENT TYPES OF SKIN DISEASES USING RASPBERRY PI (20)

More from IRJET Journal (20)

Recently uploaded (20)

DETECTION OF DIFFERENT TYPES OF SKIN DISEASES USING RASPBERRY PI