تشخیص و شناسایی ارقام سیب زمینی با ماشین بویایی بوسیله روش‌های شبکه عصبی مصنوعی و LDA

Introduction
Potato is an important vegetable that grows all over the world and is considered an important product in developing and developed countries for the human diet as a source of carbohydrates, proteins, and vitamins. This product is native to South America and its origin is from Peru, after wheat, rice and corn, it is the fourth product in the food basket of human societies. According to the statistics of the Food and Agriculture Organization of the United Nations, the area under cultivation of this crop in Iran in 2017 was 161 thousand hectares and the crop harvested from this area is about 5.1 million tons. Traditional methods of determining potato varieties were based more on morphological features, but with the production of new products, there was a need for methods that were faster and more recognizable. In the meantime, the high-performance artificial neural network can be used to classify cultivars. Artificial neural networks can classify and detect cultivars, are flexible, and are used in most agricultural products. Therefore, the olfactory machine can have high efficiency in classifying and distinguishing cultivar, originality and storage time. The olfactory machine is a system that has a different structure and approach from other methods (image processing, neural network, etc.), is flexible and is used in most agricultural products due to the presence of odour in them.
With the rapid and rapid advancement of computer technology and sensor technology, the application of the bionic electronic nose, including a semiconductor gas-sensitive sensor and a pattern recognition system as a means of detection, offers a new method for rapid classification and digit recognition. Give. The electronic nose has also introduced a new method for classifying and detecting rough rice in a non-destructive and fast way.
Due to a large number of potato varieties and sometimes the lack of familiarity of processing units with its cultivars and also time-consuming and inaccurate in identifying different potato cultivars by experts and farmers, and the importance of identifying potato cultivars and other agricultural products in At every stage of the food industry process, it is necessary to find ways to do this accurately and quickly enough. The aim of this study was to evaluate the ability and accuracy of the electronic nose with the help of an artificial neural network to detect and differentiate several potato cultivars.

Methodology
First, potatoes in 3 different cultivars (Colombo, Milvana and Sante) were prepared from Ardabil Agricultural Research Center and kept at a temperature of 10-4 ° C. One day after the data were collected, data collection began with an olfactory machine. 3-4 potatoes from each cultivar were placed in the sample container for 1 day to saturate the sample container with the smell. Then the sample chamber was connected to the electronic nasal device and data collection was performed. The data were collected by the olfactory machine in such a way that first clean air was passed through the sensor chamber for 100 seconds to clean the sensors from other odours. The odour (gases emitted from the sample) was then pumped out of the sample chamber by the pump for 100 seconds and directed to the sensors. Finally, clean air was injected into the sensor chamber for 100 seconds to prepare it for further data collection. According to these steps, the output voltage of the sensors was changed due to exposure to various gases (potato odour) and their olfactory response was collected by data collection cards, sensor signals were recorded and stored in the USB gate of the computer at 1-second intervals. A fractional method was used to correct the baseline in which noise or possible deviations were eliminated and the sensor responses were normalized and dimensionless. In the next step, linear diagnostic analysis (LDA) and artificial neural networks (ANN) were used to classify the 3 potato cultivars. LDA is a supervised method used to find the most distinctive special vectors, maximizing the ratio of the variance between class and within the class, and being able to classify two or more groups of samples. ANN and pattern recognition were used to find similarities and differences in the classification of potato cultivars. For this, 1 neuron was considered for the input layer, the hidden layer with the optimal number of neurons will be considered and five output neurons with Depending on the number of output classes the target will be considered. In-network training, logarithmic sigmoid transfer function and Lunberg-Marquardt learning method were used. Also, the amount of error was calculated using the mean square error. For learning (70%), testing (15%) and validation (15%) all data were randomly selected. Training data was provided to the network during the training and the network was adjusted according to their error. Validation was used to measure network generalization and completion of training. Data testing had no effect on training and therefore provided an independent measurement of network performance during and after training.

Conclusion
LDA and ANN methods were used to detect potato cultivars based on sensor output response. The LDA method can extract multi-sensor information to optimize resolution between classes. Therefore, this method was used to detect 3 potato cultivars based on the output response of the sensors. Detection results of cultivars equal to 100% were obtained (Figure 1). Also, in the ANN method, 8 sensors were considered in the input layer of 8 neurons according to the output data. Also, 3 layers of neurons were considered for the output layer according to the type of cultivars. Therefore, the 3-6-8 topology had the highest accuracy for detecting potato cultivars, so the RMSE value was 0.008 and the R2 value was 99.8. There was also a very high correlation between predicted and measured data (Figure 2). In this study, a portable olfactory machine system with 8 metal oxide sensors was used to investigate the detection of potato cultivars. Chemometrics methods including LDA and ANN were used for qualitative and quantitative analysis of complex data using an electronic sensor array. LDA and ANN were able to accurately identify and classify different potato cultivars with 100% accuracy. The electronic nose has the potential to be used as a fast and non-destructive method to detect different potato cultivars. Using this method in identifying potato cultivars will be very useful for researchers to select and produce pure cultivars and for farmers to produce a uniform and certified crop.