DEVELOPMENT OF COMPUTER VISION CLASSIFICATION SYSTEM FOR FRUITS
Loading...
Date
2023-05-30
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ACHARYA NG RANGA AGRICULTURAL UNIVERSITY
Abstract
Agriculture is vital to our country's growth. Demand for quality fruits grows with our country's population. To increase the income of farmers classification of fruits based on maturity and defect is required. Traditionally classification is done manually which is time consuming and labor intensive. To remain competitive, farmers must only distribute high-quality fruits that requires accuracy, speed. A computer vision system is a cost-effective system and gives consistent performance, a superior speed and accurate sorting and grading of fruits. Hence, a computer vision system that can classify fruits and vegetables based on color and defects was proposed. Images are acquired and processed using high level languages like MATLAB. To make the system cheaper and more flexible, the proposed system will use low-cost cams that may not require frame grabber hardware.
A working prototype hardware model of conveyor was developed with PC was designed and implemented to analyze the fruit quality. The fruit to be tested is placed on the circular slab and the USB camera captures and sent to the PC for classification. The QHM PC Camera features 6 white lights and has an image resolution of 25 megapixels. It has properties for adjusting image control, color saturation, brightness, sharpness and brightness. A round plastic circular slab was designed with 10 cm diameter and 4 cm height that performs the rotation and tilting operation. Open close gate was fabricated with a metal sheet of length 9 cm and width of 8 cm. Gate opens when the detected fruit is bad or immature and rejects the fruit from the conveyor. This door closes when the detected fruit is good or mature and travels along the conveyor. Actuating and control mechanism was developed which consists of Arduino, PIC and relay units. Arduino was used to transfer data between PC and PIC MCU. The Arduino is connected to the PIC microcontroller to send the signal information obtained from the PC. Crystal oscillators are connected to this PIC microcontroller to synchronize all the internal operations of the microcontroller. PIC is interfaced with LCD. A total of 5 relays are used and interfaced with PIC MCU. Signals are passed through the PIC to relay unit, the relay operates accordingly. The entire relay unit is operated on 12 V power supply.
The machine learning algorithm is used for inspecting the fruit quality whether it is good or bad as well as mature or immature, five different types of fruits such as guava, lemon, orange, apple and pomegranate have been taken for analyzing the quality.
xviii
In terms of color and defective five machine learning classifiers used such as KNN, DT, RF, ANN and SVM. The wavelet transform was used for feature extraction for fruits. All the algorithm's performance such as confusion matrix, accuracy, sensitivity, specificity, f1-score, and AUC values were calculated for all five fruits and those values were plotted. The best algorithm has been selected based on the performance and use for day 1, day 3 and day 5 storage analysis. Based on the performance SVM was found to outperform all the classifiers. Hence, SVM was used to evaluate its classification efficiency with storage conditions. The results indicated that SVM classified better on all days of storage without any misclassification. The response time of the system is 60 seconds which is low compared to other systems. It could be concluded that developed system is very suitable and useful for small-scale industries and farmers to grow up their businesses.
Key Words: Computer vision, SVM, ANN, Random Forest, Decision Tree, KNN
Description
DEVELOPMENT OF COMPUTER VISION CLASSIFICATION SYSTEM FOR FRUITS