Fruits Category Identification and Infected part Detection Using Deep Learning

Abstract

Agriculture Sector is responsible for providing a living or sustainable income to the majority of the population in the world. In Pakistan, agricultural plays a very vital role for the development and stability of the economy. Agriculture is the major source of providing food to every house as well as raw material for the development of industrial sector. Approximately 48 percent of the labor force is directly engaged with this sector. Agriculture provides numerous advantages to the population as well as it is affected by various aspects which eventually leads in effecting the overall sectors yield as well as the economy of the country on bigger scale. Various diseases in the crop is one of the leading factors which is responsible for the destruction of overall yield and the economy and it is very difficult for a naked eye to keep information of every type of crop disease and identifying the infected part of the crop. It is so important to take necessary measures for the automated solution to this problem so we could produce good quality yield and benefit the overall economy. Computer Vision and Deep Learning is one of the most utilized recent technological fields in every domain of daily life. With the advancement in computer with high quality processing these fields have made a big name in wide area of applications. We have seen so much applications of these fields in agriculture sector also for example, Crop Disease Identification, Disease Detection, Disease Area Detection etc. Many algorithms including YOLO had reviewed for this study. This thesis proposes a DL method for the classification of the fresh and rotten fruits and the category of the fruit using the CNN. In this study we have also detected the rotten part of the fruit using the YOLOV5. Dataset was calculated by two online available datasets “Fruits 360” and “Fruits fresh and rotten for classification”. Two classes of fruits from “Fruits 360” dataset were added in the “Fruits fresh and rotten for classification” dataset. Similarly, the images of two classes of fruits were captured viii through mobile phone and added in the “Fruits fresh and rotten for classification” dataset. The complete dataset contains 18k images approximately. For, detecting the rotten part of the fruit, some of the images from all the rotten fruits classes were taken and annotated to apply YOLOV5. The dataset for YOLOV5 contains 1000 annotated images approximately. Our model has achieved best possible results for classification of various categories and detection of rotten part of the fruits.