Thumbnail Image

Theses

Browse

20211
Reset filters
Subject: Farm, Machinery and Power × Author: KAU × Author: Meera, T × Start date: 2020 ×
Reset filters

Search Results

Now showing 1 - 1 of 1
  • Thumbnail Image
    ThesisItemOpen Access
    Development of a machine vision system to identify matured pepper spikes
    (Department of Farm Machinery and Power Engineering, Kelappaji College of Agricultural Engineering and Technology, Tavanur, 2021) Meera, T; KAU; Sindhu, Bhaskar
    Black pepper is a perennial crop and one of the most economically significant spices in India. It has a high commercial value in the market all around the world. Its fruit is harvested, dried and powdered for many cuisines and processed for many value added products. Black pepper is a flowering vine growing up to 4 m in height. The berries turns from green to red on maturity and are harvested when it starts to turn red. For achieving good quality and good sized pepper, it should be harvested at its proper matured state. Farmers for their time saving and due to heavy work intensity, harvest almost all the fruits which are in a range of maturity along with the real matured ones. This eventually affects the crop yield and quality. Hence employing an automated identification system in this case would be effective. An application programme interface was developed for this, using the fruit features like the shape, colour and size. By using the machine learning techniques and computer vision technology, two programmes were developed in python language, one using OpenCV library and Haar Cascade classifier, and other platform with TensorFlow as library and faster-RCNN as classifier. Studies were also carried out to analyse the physical properties of black pepper. Using image acquisition, a dataset was created and was used for training and preparation of both the models. The hardware part of the system comprised of a webcam as sensor, Raspberry Pi processor, a RPI display unit and some accessory parts. The hardware and software parts were installed and assembled, and subjected to performance evaluation. It was revealed that the Tf-RCNN platform had better performance and efficiency. The performance evaluation parameters viz., sensitivity, specificity and accuracy values were 78%, 71% and 75% respectively for the second model. It was statistically verified that there is a significant difference between the two platforms and the second model had better consistency.