Kalyan Kumar Jena1,2,3*, Sasmita Mishra2, Sarojananda Mishra2 and Sourav Kumar Bhoi3
1 Utkal University, Bhubaneswar, India
2 Department of Computer Science Engineering and Applications, Indira Gandhi Institute of Technology, Sarang, India
3 Department of Computer Science and Engineering, Parala Maharaja Engineering College, Berhampur, India
Abstract
Detecting pests in stored grain (SG) accurately is a major issue in the current scenario. It is very much essential to monitor the SG in order to take preventive measures to cease the further growth of pests in the SG. This can be done by capturing the images of SG with the help of UAVs, high-definition drones, cameras, sensors, and so on. Many methods have been introduced to detect the pests in the SG. However, no method is fully efficient in each and every situation. In this chapter, a UAV-assisted pest detection model is proposed in order to track the pests in the SG. This proposed model consists of four phases, such as data acquisition, edge detection (ED), feature extraction, and pest identification. In this model, we have only focused on the ED part by analyzing the data (pest in the SG images). Many standard ED (SED) methods, such as Sobel, Prewitt, Roberts, Morphological, Laplacian of Gaussian (LoG), Canny, are used to track the shape, location, and quantity of pests in SG. The implementation of the methods are performed using MATLAB R2015b and evaluated using signal to noise ratio (SNR), peak SNR (PSNR), and processing time (PT).
Keywords: SG, pests, UAV, SED, SNR, PSNR, PT
Grain plays an important role for the survival of human society. It is considered as one of the important basic needs of human society. Human without grain is like bike without fuel. Stored grain (SG) is a major concern in the today’s era. Several reasons are there for the loss of SG. The loss of SG due to pests is one of them. Loss of SG leads to food crisis in a global scale. Everyone has to be alert about this issue, otherwise it leads to a very serious problem to the human society globally. About 5% to 10% and 20% loss of SG by pests are caused in developed countries and developing counties, respectively [1]. So, it is essential to track the shape, location, and quantity of pests in the SG in order to take preventive measures for their growth. The SG pests can be broadly classified as primary SG pests and secondary SG pests [80, 81]. Several primary SG pests are lesser grain borer (LGB), rice weevil (RW), granary weevil (GW), angoumois grain moth (AGM), and so on, and several secondary SG pests are rust red flour beetle (RRFB), confused flour beetle (CFB), saw toothed grain beetle (STGB), flat grain beetle (FGB), warehouse moth (WM), Indian meal moth (IMM), and so on. Pest classification is shown in Figure 3.1.
In this work, a UAV [38, 39]-assisted pest detection model is proposed to monitor the SG pests by capturing the images of SG periodically. If any pest is found in the SG, these images will be sent to the information center (IC) and then these images will be collected from the IC for processing and analysis. The captured images will be processed by the help of standard edge detection (SED) methods, such as Sobel [40–43], Prewitt [44–46], Roberts [47–49], Morphological [54], LoG [50–53], Canny [55–58], and several other image processing approaches [59–79] to track the shape, location, and quantity of pests in the SG [1–37]. The outputs of SED methods are compared and evaluated using signal to noise ratio (SNR), peak SNR (PSNR), and processing time (PT).
The main contributions in this chapter are stated as follows:
The rest of the chapter is organized as follows: Sections II, III, IV, and V describe the related works, proposed model, results and discussion, and conclusion of the work, respectively.
Several works have proposed by several researchers to identify the pests in the SG [1–37]. In this section, some of the works were discussed. Qin et al. [1] focuses on spectral residual (SR) saliency ED mechanism to detect the pests in SG. It mainly focuses on the image logarithmic spectrum as part of the information of the image. The rest of the spectrum is changed to airspace for obtaining the results (ED). It is based on the processing in frequency domain. Shen et al. [2] proposed a deep neural network-based technique to identify and detect the insects in SG. In this work, faster R-CNN is used to focus the regions of insects in the images as well as to classify the insects in the regions. An advanced inception network can be developed for feature map extraction. Solà et al. [3] proposed a multiplex PCR method to identify and detect the pests, such as Sitotroga cerealella, S. oryzae, Sitophilus granarius, Rhyzopertha dominica, and S. zeamais, which are hidden in grain kernels. This method may be used for decision making (commercial) to satisfy the demands in the current market scenario.
Kaushik and Singhai [4] proposed an integrated and environment monitoring-sensing system for the detection of contamination and insect infestation in SG. This method may monitor the quality of grain. The sensor data may help in the prediction of essential information and provide alerts for taking preventive actions. Liu et al. [6] proposed a deep learning-based pestnet method to detect and classify multiclass pests. The pestnet method focuses on CSA, Convolutional Neural Network (CNN), Region Proposal Network (RPN), Position-Sensitive Score Map (PSSM), Fully Connected (FC), as well as contextual Region of Interest (RoI) mechanisms. The pestnet method can be evaluated by using MPD2018 data set as well as newly referred pest image data set. Priyadarsini et al. [7] focuses on a device (smart) to explore the negative effects on the pest bugs in case of land (harvesting), as well as in the water body PH level. It is mainly focused on the identification of the PH value and the control of pest for better agriculture. In order to repel several types of insects, different ultrasonic waves are produced.
The proposed model is mentioned in Figure 3.2. This model is focused on UAV-assisted pest detection model for the detection of pests over the SG. It is very essential to track each and every pest in the SG in order to take preventive measures for further loss of SG. The UAVs or high-definition drones can be used to monitor the SG at regular intervals. The UAVs will capture the SG images at regular intervals and send these images to the IC. The captured images can be taken from the IC for analysis periodically. If any pest is detected, then the captured images will be processed immediately in order to track the position, shape, and quantity of pests. It can be done by using four phases discussed as follows.
In this work, several SG pest images with different sizes, such as RW, LGB, RRFB, CFB, GW, are taken from source images [80, 81] which are mentioned in Figures 3.3 to 3.7. MATLAB R2015b is used for the processing of SG pest images. The SG pest images are processed using several SED methods, such as Sobel, Prewitt, Roberts, Morphological, LoG, Canny methods and evaluated using SNR, PSNR, and PT. The quality of the output image increases if the PSNR or SNR value increases. The method performs faster if it deals with lesser PT.
From the analysis of Figures 3.3 to 3.10 and Tables 3.1 to 3.3, it is concluded that the morphological method detects the pests in a better way as compared with other methods, and it processes the images with less PT. However, the PSNR and SNR values of Canny method is higher as compared with others. From the analysis of RW image as mentioned in Figure 3.3, morphological method detects the RW from image in a better way, and its PT is 0.01 unit. However, Sobel, Prewitt, and Roberts method try to detect the RW. However, the results of these methods are not so good as compared with the morphological method. The LoG and Canny methods are not able to identify the RW. From the analysis of LGB and RRFB images, as mentioned in Figures 3.4 and 3.5, respectively, morphological method detects the LGB and RRFB pests from images in a better way as compared with other methods, and its PT are 0.03 and 0.01 units, respectively. However, Sobel, Prewitt, and Roberts methods try to detect the LGB and RRFB pests, and LoG and Canny methods also try to detect the LGB and RRFB from the images. However, the results of LoG and Canny methods are not so good as compared with Sobel, Prewitt, and Roberts methods.
Table 3.1 Evaluation of methods using SNR(dB) value.
Method | RW | LGB | RRFB | CFB | GW |
Sobel | 13.87 | 14.15 | 15.29 | 14.59 | 14.55 |
Prewitt | 13.87 | 14.15 | 15.30 | 14.60 | 14.55 |
Roberts | 13.84 | 14.14 | 15.24 | 14.51 | 14.46 |
LoG | 14.23 | 14.87 | 15.69 | 15.25 | 14.89 |
Canny | 14.62 | 15.95 | 16.34 | 16.11 | 15.72 |
Morphological | 13.75 | 14.16 | 15.36 | 14.71 | 14.78 |
Method | RW | LGB | RRFB | CFB | GW |
Sobel | 19.09 | 19.18 | 19.55 | 19. 32 | 19.29 |
Prewitt | 19.09 | 19.18 | 19.57 | 19.32 | 19.29 |
Roberts | 19.09 | 19.17 | 19.54 | 19.29 | 19.27 |
LoG | 19.20 | 19.39 | 19.69 | 19.52 | 19.41 |
Canny | 19.32 | 19.78 | 19.95 | 19.58 | 19.70 |
Morphological | 19.07 | 19.18 | 19.58 | 19.37 | 19.37 |
Table 3.3 PT (in units) calculation of several methods.
Method | RW | LGB | RRFB | CFB | GW |
Sobel | 0.08 | 0.06 | 0.01 | 0.03 | 0.03 |
Prewitt | 0.08 | 0.05 | 0.01 | 0.02 | 0.02 |
Roberts | 0.12 | 0.05 | 0.01 | 0.01 | 0.03 |
LoG | 0.27 | 0.05 | 0.03 | 0.05 | 0.14 |
Canny | 0.11 | 0.05 | 0.08 | 0.12 | 0.03 |
Morphological | 0.01 | 0.03 | 0.01 | 0.01 | 0.01 |
Similarly, from the analysis of CFB and GW images, as mentioned in Figures 3.6 and 3.7, respectively, morphological method detects the CFB and GW pests from images in a better way as compared with others, and its PT is 0.01 unit in both cases. However, Sobel, Prewitt, and Roberts methods try to detect the CFB and GW, and LoG and Canny methods also try to detect the CFB and GW pests from the images. However, the results of LoG and Canny methods are not so good as compared with the Sobel, Prewitt, and Roberts methods.
In this chapter, a UAV-assisted pest detection model is proposed which mainly consists of four phases, such as data acquisition, ED, feature extraction, and pest identification. In this work, we have only focused on the ED part by analyzing the pests in the SG images. The implementation of the SED methods is performed using MATLAB R2015b and evaluated using SNR, PSNR, and PT. From the results, it is observed that morphological ED method detects the edges well with lower PT. In the future, this SED method will be taken in our pest detection model. Then, the edge detected images are given to a classifier for learning to identify the pests accurately. This proposed model will be a better solution for the prevention of SGs.
18.222.164.141