Real-time Visual Identification of Defect, Size, Maturity, and Quality on Ladies' Finger, Bitter Gourd, and Cucumber Using Image Processing and MobileNetV2

Agriculture makes substantial use of image processing. Technology's non-destructive nature is one of its benefits. For smart farming and smart agriculture, numerous image-processing techniques are applied, including categorization, detection, grading, etc. Using a deep learning system that is growing closer to human-level recognition, the task of defect, size, maturity, and quality detection on ladies' fingers, bitter gourds, and cucumbers is challenging. The proposed study will benefit farmers by automatically classifying the defect, size, maturity, and quality of the ladies' finger, bitter gourd, and cucumber planted on the farm before its harvest. This study aims to detect the defect, size, maturity, and quality of the ladies' finger, bitter gourd, and cucumber using Watershed Algorithm, Otsu Thresholding, RGB Masking, K-means Cluster Algorithm, and MobileNetV2. This study had an overall accuracy of 90.59, 91.74, 91.83, and 91.45 percentages for the defect, size, maturity, and quality identification of ladies' fingers. Meanwhile, the bitter gourd has an accuracy of 90.78, 93.07, 90.57, and 90.45 percentages for identifying the defect, size, maturity, and quality, respectively. Lastly, for the cucumber, the accuracy value for defect identification is 89 percent, size identification is 91.43 percent, maturity identification is 90.79 percent, and quality identification is 90.95 percent.