You are on page 1of 13

INTRODUCTION

Problem Definition:

❑ This project is representing a mobile app detecting


recyclable and nonrecyclable materials from an
image.

❑ Live pictures can be taken from a video being


captured.

❑ The distinct constituents of the image can be


identified.

❑ Difference of constituents according to the


features is depicted as well.
MOTIVATION:

✔ To make use of the recyclable materials depicted from the image


processing.

✔ Leading to sustainable growth of useful resources.

✔ Mobile app being an economical resource due to it’s cost effectiveness.

✔ High-efficiency and time management.

2
OBJECTIVE:

⮚ Reuse of efficient resources from the garbage dump to make a cleaner


environment.

⮚ Differentiating among ferrous metals and non-ferrous metals for further


processing of distinct materials.

⮚ Using CNN ,recognition of smaller local patterns on the image is being


measured and concatenate them into more complex shapes by
generation of BLUE scores for proper accuracy.
3
Contributions:

❖ Ayan Banerjee[1] designed the front end portion of recycler


app for compatibility of Kotlin-Kotlin connection.Set up the Ip
based camera to properly identify and detect the waste images
.
❖ Suprava Saha[2] studied the research papers regarding the
aspect and also handled the Kotlin connection. She made the
final ppt for the topic .
❖ Deepika Das[3] exported the keras model made in cnn to tflite
and enabled the tflite model in android studio to view the
results in our current app. She also designed some aspect of
4
image classification.
TO BE CONTD….
❖ Soham Chatterjee[4] wrote the image processing
code in keras and handled the captioning part of
the detected images .Slightly helped writing the ppt
along with [2].Exported and sorted the datasets in
proper order. Enabled the conversion of the app
into its iOS equivalent.
❖ Aditya Kr. Singh [5] generated the database link
for network connection and integrated the IP
address of Ip camera into own IP address by
subnet masking (Android studio).
❖ Arkaprava Mazumder[6] presented the last screen
show and generated the capture markings for the
5 final captured images.
Research Review:

✔ Bobulski- Plastic Waste Segmentation using CNN proposes a system for classifying
waste by-polyethylene terephthalate,high-density polyethylene,etc.
✔ Mittal & Yagnik-Image is classified and depicted as garbage dump which employs CNN
called GarbNet for resource constrained environment.
✔ Jogin & Mohana- This paper depicted feature extraction using CNN and DL, features are
extracted using images to identify the specific image.
✔ Kothari & Tyagi- Renewable waste materials from different sectors are transferred into useful
energy like-biohydrogen,biogas.etc.
✔ Fredrik Almin- All non-ferrous materials are identified from the given pictures being taken as
an input using methods like-automate sorting of computer vision.

6
Data Set Preparation

• Collecting the garbage data sets from Kaggle.


• Further segmenting it into 3 sub-folders.
• We have taken 9 categories of wastes for classification : cardboard, plastic, metal,
• First, containing the training set(70%) of the dataset for training purposes.
• Second, containing the valid set (10%) of the dataset for validation purposes.
• Third, containing the testing data (20%) of the dataset for testing purposes.

7
CATEGORY OF TRAIN _DATA (NO OF TEST_DATA(NO OF VALID_DATA(NO OF TECHNICAL STACK
WASTES IMAGES) IMAGES) IMAGES) IMPLEMENTATION AND
IMPORTED

CARDBOARD 613 191 87 CNN/KAGGLE

BATTERY 611 189 145 CNN/KAGGLE

CLOTHES 3727 1065 533 CNN/KAGGLE

GLASS 251 171 79 CNN/KAGGLE

METAL 500 169 100 CNN/KAGGLE

PAPER 700 250 100 CNN/KAGGLE

PLASTIC 605 173 87 CNN/KAGGLE

SHOES 1384 395 198 CNN/KAGGLE

TRASH 488 139 70 CNN/KAGGLE


8
Methodology:

Step 1: Image capturing: Capturing is being done both using normal camera as well as IP
camera and then sub-folders are being created for the same.

Step 2:Filtration of image: These subfolders are being created as training set, validation set
and test set for further classification of data.

Step 3:Image Identification using CNN(DL): At last, final processing is being made and the
identified recyclable materials has been identified for further processing.
Step 4: After image preprocessing, we have generate captions for the detected images and
generate BLUE scores for better accuracy.
9 Step 5: Generate further results for detection of the selected images.
EXPLANATION WITH FLOWCHART

10
Results:

11
Conclusion:

✔ From the garbage materials being collected from the mines, we can identify the
recyclable metals and further retrieve it for further processing.

✔ Therefore, the sustainable resources can be recovered and thus it helps in


making a recyclable growth of nature.

✔ Thus, mining of resources can be made more resourceful using a mobile app
which ensures low cost and high efficiency.
Future Scope

▪ GPS can be incorporated into the system for further processing of the system.
▪ Therefore, mapping to the unit from the device location can be accessed.
▪ Which shop is recycling which material can also be accessed for the same.
▪ Thereby, unnecessary mining garbage dump can be avoided to a large extent.

13

You might also like