الفهرس 
REVIEW OF LITERATUREOnly 14 pages are availabe for public view
 |  | from | 126 |  |  |
| | | from | 126 | | |
Abstract
SUMMARY AND CONCLUSION
Through studying some of the physical properties of this study samples, and studying classification dates with convolutional neural network, the following were shown:
• Some difference appears between the physical properties of the accepted and rejected dates. In the Mejdool variety there is a difference between the mean for the properties of length, average width, mass, and volume. According to the study sample, it was shown that the fruits of dates are more than (26 mm, 22 gm and 19 cm3) for the physical properties (average width, mass, and size), respectively.
• Generally, as a result of the existence of an overlap in the range between the physical properties under study of the accepted and rejected date fruits, which reduces reliance on them in the classification processes.
• Under experimental conditions, the use of thermal imaging is not sufficient to classify accepted dates from the rejected dates, so another method of classification has been resorted to identify the different features.
• Some of Convolutional neural network architectures (Inception V3, Inception Resnet V2, and VGG 19) used to evaluate the date health state, and to study which model will achieve high performance for classification process.
• Building dataset for the accepted and rejected dates for the three selected varieties (Mejdool, El- wadi, Saeidi) with total number of images 1484 image for accepted and 2803 image for the rejected. In addition to an open source (Kaggle) dataset was added to accepted to balance the data.
• Among the above results it showed that the highest-performance result for classifying dates is Inception Resnet V2 model with highest accuracy 98.99% and lowest loss 0.0344 at test stage.
• Finally for a practical test with real dataset which differ from the input dataset, Inception Resnet V2 identified 82.5% percentage correctly out of 40 random samples.