Evaluation and Segregation of Fruit Quality using Machine and Deep Learning Techniques

Identifying fruit fruits is an essential part of fruit plantation smart management. This paper presents a mechanism based on the available deep learning model to determine the fruit fast and reliably in a complicated orchard environment. We employed the YOLOv3 method to detect the deep characteristi...

Full description

Saved in:

Bibliographic Details
Published in:	2022 International Conference on Futuristic Technologies (INCOFT) pp. 1 - 8
Main Authors:	Mali, Mohit Kedar, Devake, Shubham Rajendra, Kharpude, Satyam Manoj, Kumar, Yogesh, Pal, Prashant, Singh, Shashank Kumar, Bansod, Saurabh
Format:	Conference Proceeding
Language:	English
Published:	IEEE 25-11-2022
Subjects:	Cameras Data preprocessing Deep learning Fruit Detection Image edge detection image processing Image segmentation Object detection Real-time systems Segmentation U-Net VGG16 Yolov3 model
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Identifying fruit fruits is an essential part of fruit plantation smart management. This paper presents a mechanism based on the available deep learning model to determine the fruit fast and reliably in a complicated orchard environment. We employed the YOLOv3 method to detect the deep characteristics of fruit fruits using a stereo camera and an indoor fruit dataset, resulting in efficient identification of varied fruit sizes. In this paper, we proposed segmentation of fruit using the UNET framework using various deep learning backbones such as densenet, efficientnet, mobilenet, vggnet etc. This research has been done on field fruit images, so it contains some noise in the background of the image; this problem impacts accurate detection and classification. The YOLOv3 model has been used for fruit object detection with boundary regions. It returns the normalized image in square form with the detection of edges of the object the UNET framework has used for classification. The various backbones obtain different accuracy, but UNET-VGG19 brings Dice Coefficient of 90.35%, which is better than other methods. As a result, various experimental analyses were done on real-time images and demonstrated the dice score, precision, recall, etc.
DOI:	10.1109/INCOFT55651.2022.10094447