Potato is the world’s fourth-largest food crop and has been defined as “the food of the future”. Developing the potato industry is of great significance to ensure food safety and for improving agricultural economics [1,2]. Potato soft rot is a disease which is composed of several types of either erwinia alone or as a mixed infection. This bacterial disease can harm potato tuber during the storage period. The severiety of occurrance is different degree every year and has become one of the major diseases of potato throughout the world. This type of potato bacterial disease can occurr at the seedling or storage stage thus infecting the surrounding healthy potato tubers quickly and easily which may lead to severe economic losses [3,4]. Sensitive and rapid disease detection systems are required for disease prevention and control. The gram staining method , physiological and biochemical method , enzyme-linked adsorption measurement method , and the based PCR detection technique  are currently used to detect and identify bacterial agents responsible for the disease. However, the disease degree of the potato cannot be detected in real-time with accuracy and reliability . To improve the performance of gas concentration detection of diseased potatoes, and to better meet the requirements of practical applications, many experts and scholars have dedicated resources for the study by using an electronic nose to obtain different smell fingerprint data via a test of potato diseases. Relative to healthy potatoes, the most obvious feature is that volatile components obviously change during potato tuber soft rot disease and the smell also reveals significant differences. Therefore, there is a good correlation between the smell of potatoes and its degree of damage. The potato disease can be detected firstly by its volatile gas concentration and thus, the disease degree can be recognized [9,10,11,12,14,15]. Stinson et al. proposed a healthy recognition ability of the disease in potato by an electronic nose ; however, the technology required further research; DE Lacy Costello et al. designed an electronic nose system, which could detect diseased potatoes within healthy potatoes ; however, the recognition accuracy required further improvement. The PCA and LDA arithmetic were used for data analysis by Biondi’s team to detect potato diseases based on the electronic nose; however, the LDA recognition rate was slightly lower under active sampling condition . This reveals that although the conventional electronic nose detection technology has already been developed, shortages remained and it was difficult to meet the requirements of potato soft rot bacteria invasion detection. Based on the current research situation of the potato disease detection by an electronic nose system, the bionic optimization design of electronic nose gas chamber test system was used to detect volatile organic compounds at varying concentrations of the potato tuber soft rot disease, and the disease degree of the potato tuber was studied and different algorithms were used to verify the validity and accuracy of the bionic electronic nose system.
2 Materials and methods
2.1 Preparation and inoculation of bacteria to potato tubers
Potato soft rot is mainly caused by the soft rot erwinia pathogen of carrots soft rot. Carrots soft rot erwinia potato consists of black shank subspecies and chrysanthemum Euclidean bacteria. The carrot soft inoculation of the pathogen Euclidean coli carrots subspecies (Ecc) was selected for this experiment, and after choosing the Ecc it was placed into potato dextrose agar (PDA) medium. Before use, the pathogen was inoculated on PDA medium at room temperature for activation, and the bacterial suspension was made using sterile distilled water. Healthy, disease-free potatoes, without damage on the surface and similar in size were obtained and cleaned with clear water, and then dried. The lenticels inoculation method  was adopted, the bacterial suspension was used, and the potato tubers were infected via vaccination.
2.2 Orthogonal Test
The best detected conditions of electronic nose to the diseased potato were determined by the orthogonal experimental method. According to the four factors of sample weight (A), the tuber diameter size (B), sampling time (C), and sampling interval (D), orthogonal experimental factors level table (L9 (34)) was established, as shown in Table 1. The maximum response value of the electronic nose was chosen as detecting indicators. Variance analysis and factor significance test were conducted via SPSS statistical software.
The variance results are shown in Table 2. It can be seen that the factor B with the largest range, and the influence degree of the factors to the maximum response of electronic nose detection was B > C > A > D. The optimum detection combination condition was A3B1C2D2 which was obtained via comparison of different testing conditions. The potato tuber weighing 200 g, diameter of1 cm, sampling time, and sampling interval are 120 s and 3 min at each test, respectively. Each test was repeated thrice and the average from the three tests is reported as the result.
2.3 Test apparatus
2.3.1 Experimental apparatus
The instruments, which were used for this experiment, are the bionic electronic nose system, a sample bottle, a data collection analyzer, and a volatile gas-sampling device.
2.3.2 The potato volatile gas-sampling device
The sampling device of the potato gas volatiles is shown in Figure 1. The system is composed of sampling needle, cork, trachea, potato samples, glass tube of gas collecting system, plastic wrap, and an air filter (containing activated carbon).
2.3.3 Bionic electronic nose sensor array and gas chamber optimization
A sensor array is the core component of a bionic electronic nose system. The utilized detection principle converts the smell of the gas molecule signals into a voltage signal via the cross sensitivity of the sensor. Since the gas sensitive material is different, many different types of gas sensor exist and metal oxide semiconductor gas sensors are the most commonly used. Potato tubers can induce the release of ethanol, acetone, 2-butanone, and the other specific components following bacterial infection with soft rot . According to the experiment and the characteristics of gas composition, the TGS series gas sensor (FIGARO company, Japan) was selected to compose the sensor array for the initial experimental research, and the related specific parameters of sensor are shown in Table 3.
Based on the adult structural features of the nasal cavity and the nasal cavity internal flow characteristics of fluid mechanics research, a bionic gas chamber system was designed. The geometry of this gas chamber featured the shape of a cone to emulate a big cavity and small jaws of gas chamber structure, to increase the inlet air flow rate compared to the air velocity of the inner cavity; thus providing sufficient material for contact with the sensing element. The specific shape, volume, and wind speed of the fan of the conical bionic gas chamber were determined via orthogonal testing . The jaw radius of the electronic nose gas chamber was r = 50 mm, the cavity radius of the electronic nose gas chamber was R = 120 mm, and the height of the gas chamber was H = 200 mm. Different types of wires of sensors were arranged to avoid the flow channels and to fix the position of the sensor array in boxes export origin. Thus the power cord moved from the outside into the data line. The draft fan was installed within the external sensor array to guide the gas flow and the wind speed was set to V = 1.5 m/s. The electronic nose array and the gas chamber system are shown in Figure 2.
2.4 Experimental method
Ten groups of controlled experiments were initiated, respectively marked as 1, 2 … to 10. Two sample bottles were placed in each group, labeled A and B, respectively. One piece of infected tuber and nine pieces of disease-free tubers were placed in bottle A, while ten pieces of disease-free tubers were placed in bottle B. According to disease severity the tubers involved in the area ratio of the disease were divided into six levels. Classification standards are shown in Table 4 as follows:
Experimental tests were carried out using the electronic nose system. During the test, the indoor air should be kept fresh and clean, the indoor temperature was 22 ± 1°C, and the humidity was 60% ± 1%. At the beginning of each test, preheating and 0.5 hour of air purification were required; Due to the different temperature condition, the volatile gas composition of potato tube were different. So, the effect of temperature factor on the volatile component must be considered. Different temperature cases of potato tuber volatile component were collected and analyzed. The results showed that the volatile components were less below 10°C, and they were complex above 40°C and will affect the accuracy of the sensors, and volatile components were moderate between 20°C to 40°C. Highest volatile component were observed at 30°C and the potato basic physiological state could be guaranteed relatively constant. So, the potato tuber samples were heated to 30°C using a heating box, and then they were taken for gas collecting indoor and covered with plastic wrap; the signal acquisition was done after the sensor reached a steady state. During acquisition, both sampling needles were inserted and set at the indoor gas. The cork of the sampling device was inserted into the inlet port of the electronic nose system for data collection and recording. Sampling time was set to 2 min and when the collection was completed, the chamber room was filled with clean air to clean the sensor reaction. Then, the gas reduction time was set to 3 min and the next set of experiments were carried out after gas cleaning. Measurements were taken from each group daily for seven days generating 140 data points. 10-fold cross validation was used for training and testing. Artificial neural network (RBF NN) and support vector machine (SVM) were utilized for performance analysis, and the difference of the estimation and the system availability were validated [21,22]. An artificial neural network is widely used in pattern recognition due to its strong nonlinear mapping ability and superior fault tolerance. A three-layer RBF neural network structure was constructed in this study via the RBF neural network algorithm. Six input layer nodes, three hidden layer nodes, and one output layer node were utilized. A Gaussian function was chosen as the neuron function in a hidden layer of the RBF neural network. For the hidden layer center of the Gaussian function, the orthogonal least squares method was chosen for this study, and the weights of the network output were trained via the least square method with the training goal of minimizing total error. The training samples of the RBF neural network were potato tuber samples with different degree of soft rot disease infection. The radial basis function of good classification performance and stability was selected in the SVM algorithm.
The conducted research is not related to either human or animals use
3.1 Response of sensor arrays to volatile organic compounds in potato
The infected potato tuber and healthy potato tuber electronic nose detection were carried out by the dynamic headspace sampling method. The electronic nose sensor response curves on diseased potato tubers and disease-free potato tubers are shown in Figure 3. The x-axis shows time, the vertical axis shows the signal value, and signal values show the output voltage value of each sensor. It can be clearly seen from Figure 3 that the output voltage value of each sensor changed slightly in the beginning stages. As the volatiles attached on the surface of the sensor, the sensor output voltage value changed constantly, and then flattened out, and finally reached a stable state. In addition, the response curves of each sensor on diseased potato tuber (Figure 3a) and disease-free potato tuber were different (Figure 3b). This difference indicates that the electronic nose sensor array system is able to detect the potato diseases.
The sensor output average curve graph of the different disease degree level is shown in Figure 4. As can be seen from Figure 4, infected potatoes began to emit volatile compounds with the increase of the level of infection, and the volatile compounds accumulated and increased. There is a good correlation between the potato smell and its degree of damage as the output voltage value of the sensor increased. But when disease degree increased to level 5, potato prevalence rate reached 70% to 100%, the volatile component was more complex when compared with other conditions. The complex component maybe attributed to the decrease in the response between level 4 to 5. Figure 4 also suggest that the output voltage curves of the five sensors were different, and that the selected sensors have a very good sensitivity.
3.2 Feature extraction and optimization
According to the characteristics of the sampled smell, the maximum gradient, average gradient and the minimum gradient were picked up as characteristic values, and they were recorded as Max, Average, and Min, respectively. Fifteen characteristic values were obtained via five sensors and a matrix was built from these. The column of the eigenvalues matrix was taken as a characteristic vector and was utilized as the input of the RBF neural network and SVM for training and testing. The recognition rates are shown in Table 5. The table reveals that the maximum value of outstanding characteristics, while the characteristics of the mean value and maximum gradient values are not outstanding; therefore, the maximum vector was defined as a feature vector.
3.3 RBF NN and SVM comparison
A total of 140 groups of data were measured in these tests. The data was divided into 10 groups, with 14 sets of data in each group, and the data was processed by 10-fold cross validation method , in the method, data were divided to ten groups of which nine groups were set as training sets and the other one group was set as test set in turn, the average of 10 results was taken as the estimate of the recognition and the evaluation measurements can be represented by an average value with a standard deviation. The influence of sample sizes to the identifying rate was compared, and the stability difference of the recognition rate was obtained between the RBF network and SVM recognition. The results and the number of samples of each level are shown in Table 6.
In Table 6, when the disease level was level 0, the RBF recognition rate is 73.1%, the SVM recognition rate is 78.6%, and the SVM algorithm recognition rate is 5.5% higher than that of the RBF algorithm. When the disease level is between 1 to 5, the SVM algorithm recognition rate are 4.4%, 3.5%, 2.9%, 2.2% and 3.6% higher than that of the RBF algorithm, respectively. The analysis results reveal that the SVM algorithm recognition rates are higher than that of RBF algorithm. And the early detection of potato soft rot disease is feasible using an electronic nose and model based on RBF neural network and SVM. Furthermore, the SVM algorithm is better than the RBF algorithm.
In this study, a type of potato tuber soft rot disease early detection method based on electronic nose technology was introduced. An optimized bionic electronic nose gas chamber and a scientific and reasonable sampling device was designed, and changes of volatile substances of infected soft rot disease of potato tubers were detected. The RBF neural network algorithm and the SVM algorithm were utilized to detect and identify infection with soft rot disease of potato tuber samples, and the results shown that the proposed bionic electronic nose system can be used for early detection of potato tuber soft rot disease; Via comparison and analysis, the recognition rate via SVM algorithm reached up to 89.7%, and the results were superior to the RBM NN algorithm.
This work was supported by Jilin province development and reform commission (Grant No. 2016C029), China Postdoctoral Science Foundation (Grant No. 2015M571367 and No. 2016M601383), The Education Department of Jilin Province (Grant No. [ 2015 ] 489, No. [ 2015 ] 449, No. JJKH20170796KJ and No. JJKH20170812KJ).
Luo Q.Y., Liu Y., Gao M.J., Yi X.F., Status Quo and Prospect of China’s Potato Industry, Agricultural Outlook, 2015, 3(11), 35-40 Google Scholar
Ma H., Progress on Control of Potato Soft Rot in China, Biotechnology Bulletin, 2007(1), 42-44 Google Scholar
Lui L.H., Vikram A., Abu-Nada Y., Kushalappa A.C., Raghavan G.S.V., A1-Mughrabi K., Volatile Metabolic Profiling for Discrimination of Potato Tubers Inoculated with Dry and Soft Rot Pathogens. Amer. J. of Potato Res., 2005, 82, 1-8 CrossrefGoogle Scholar
Fang Z.D., Plant disease research methods, China Agricultural Press, 1998 Google Scholar
Han G.T., Yang Z.H., Zhu J.H., Zhao D.M., Han Y.Q., Development of Duplex PCR Assay for Detection of Clavibacter michiganensis subsp sepedonicus and Pectobacterium atroseptica from Potato, Scientia Agricultura Sinica 2011, 44(20), 4199-4206 Google Scholar
Guo F.L., Zhang H.Y., Yu X.M., Zhao W.Q., Liu D.Q., Rapid PCR Detection Technology of Common Scab Pathogens in China, Scientia Agricultura Sinica, 2013, 46(23), 4926-4932 Google Scholar
Wang M., Zhang R., Test for Soft Rot Pathogenic Bacteria on Potato via the Method of 16SrDNA PCR, Chinese Potato, 2007, 21(4), 206-208 Google Scholar
de Lacy Costello B., Evans P., Ewen R.J., Gunson H., Ratcliffe N.M., Spencer-Phillips P.T., Identification of volatiles generated by potato tubers (Solanumtuberosum cv: Maris Piper) infected by Erwinia carotovora, Bacillus polymyxa and Arthrobacter sp., Plant Pathology, 1999, 48(3), 345-351 Google Scholar
Waterer D.R, Pritchard M.K., Monitoring of volatiles: A technique for detection of soft rot (Erwinia carotovora) in potato tubers, Canadian Journal of Plant Pathology, 1984, 6(2), 165-171 CrossrefGoogle Scholar
Czajkowski R., Perombelon M.C.M., Jafra S., Lojkowska E., Potrykus M., van der Wolf J.M., et al., Detection, identification and differentiation of Pectobacterium and Dickeya species causing potato blackleg and tuber soft rot: a review, Annals of Applied Biology, 2015, 166(1), 18-38 Web of ScienceCrossrefGoogle Scholar
Artur B.O, Sylvio L.H., Claudio L.M., Perla A.G., Effect of UV-C radiation and fluorescent light to control postharvest soft rot in potato seed tubers, Scientia Horticulturae, 2015, 181, 174-181CrossrefWeb of ScienceGoogle Scholar
Thabèt Y., Sami S., Abdelhafidh D., Sensitivity of Pectobacterium carotovorum to hydroxytyrosol-rich extracts and their effect on the development of soft rot in potato tubers during storage, Crop Protection, 2013, 53, 52-57 Web of ScienceCrossrefGoogle Scholar
Chatterjee D., Bhattacharjee P., Bhattacharyyab N., Development of methodology for assessment of shelf-life of fried potato wedges using electronic noses: Sensor screening by fuzzy logic analysis, Journal of Food Engineering, 2014, 133, 23-29 Web of ScienceCrossrefGoogle Scholar
de Lacy Costello B., Ewen R.J., Gunson H.E., Ratcliffe N.M., Spencer-Phillips P.T., The development of a sensor system for the early detection of soft rot in stored potato tubers. Measurement Science and Technology, 2000, 11(12), 1685. CrossrefGoogle Scholar
Biondi E., Blasioli S., Galeone A., Spinelli F., Cellini A., Lucchese C., et al., Detection of potato brown rot and ring rot by electronic nose: From laboratory to real scale, Talanta, 2014, 129, 422-430 CrossrefWeb of SciencePubMedGoogle Scholar
Chang Z.Y, Chen D.H, Men H.T., Tong J., Xie J., Electronic nose in chicken freshness detection based on cone-shaped gas chamber bionic technology, Journal of Jilin University: Eng and Technol Ed., 2011, 41(A2), 334-337 Google Scholar
Li H., Statistics and learning methods, Tsinghua university press, 2012, 95-134 Google Scholar
Zhang H., Yang S., Guo L., Zhao Y., Shao F., Chen F., Comparisons of isomiR patterns and classification performance using the rank-based MANOVA and 10-fold cross-validation, Gene, 2015, 569, 21-26 Web of SciencePubMedCrossrefGoogle Scholar
About the article
Published Online: 2017-11-23
Conflict of interest: Authors state no conflict of interest.
Citation Information: Open Life Sciences, Volume 12, Issue 1, Pages 379–385, ISSN (Online) 2391-5412, DOI: https://doi.org/10.1515/biol-2017-0044.
© 2017 Zhiyong Chang et al.. This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. BY-NC-ND 4.0