Prediction of output moisture content of dill from hot-air conveyor belt dryer using machine vision

Document Type : Research Paper

Authors

• Hawin Alipanahi ¹
• Nasser Behroozi-Khazaei ²
• Kaveh Mollazade ¹
• Hosain Darvishi ¹

¹ Department of Biosystems Engineering, Faculty of Agriculture, University of Kurdistan, Sanandaj, Iran

² Department of Biosystems Engineering, Faculty of Agriculture, University of Kurdistan, Sanandaj, Iran.

Abstract

Keywords

• Artificial neural network
• Moisture content
• Machine vision
• Shrinkage
• Image processing

Main Subjects

• drying

Prediction of output moisture content of dill from hot-air conveyor belt dryer using machine vision

Extended Abstract

Introduction

Due to the fact that high moisture content of agricultural products such as dill promotes the growth and activity of microorganisms. Therefore, they are typically dried before being introduced to the market to prevent biochemical reactions. Due to the disadvantages of traditional methods of the drying agricultural products, Hot-air conveyor belt dryers as industrial dryers are commonly used to dry of the agricultural products due to their simplicity of operation, easy access and cost-effective. But the systems or sensors that are used to correctly determine or measure the final moisture level from the output of this dryer have low accuracy or are not common due to their other disadvantages (Mizukami et al., 2006; Rywotycki, 2003). Today, machine vision and artificial neural network demonstrated their ability for online monitoring of drying processes and process modeling respectively. The application of these technologies to control the drying process are being widely developed (Li & Chen, 2020; Rezaei et al., 2019; Su et al., 2015). Suprapto & Riyanto (2020) studied the process of drying grapes in a conveyor belt dryer equipped with a machine vision system. Machine vision system was used to image the grapes during drying and neural network was used for modeling. Our research aimed to investigate the feasibility and application of machine vision in conveyor belt dryers for measuring moisture content and controlling the drying process. We found that there was a lack of studies focusing on this specific area.

Materials and Methods

Dill used in the experiments. The initial and final moisture content of products exit from the dryer was determined by AOAC (1980) method. In this study, a conveyor belt hot air dryer equipped with a machine vision system was developed (Fig. 1). This dryer consists of machine vision, conveyer belt, heater, SSR rely, temperature and belts speed control program. The machine vision system including three webcams (Logitech C920, FULL HD-1920*1080 pixels-30fps-Switzerland) were installed in the entrance, exit and middle part of the dryer (with equal distances from each other). The images taken by these webcams were transferred to the computer through a USB cable and received with the help of MATLAB 2018b software. Two LED tube lamps (230 v, power 18W, G13 base, length 1.20 m, manufactured by Pars Shahab Company, Iran) were used for lighting. The experiments was done with two levels of air temperature (50 and 60 C) and three levels of belt speed for each air temperature (Table 1).

An image processing algorithm was developed to extract features from the image. The first step to develop the desired algorithm is image segmentation to remove the background from the images (Fig. 2). After segmenting the images, shrinkage as a feature was extract following equation:

                                                                                                                     (1)

Where  the number of pixel of object with camera in the middle and exit of dryer and   is the number of pixel with camera in the entrance of the dryer. Also the shrinkage difference between cameras extracted with flowing equations:

                                                                                                                                                  (2)

                                                                                                                                        (3)

                                                                                                                                       (4)

For modeling using the neural network in this study, ∆A₁₂, ∆A₂₃, ∆A₁₃, and the initial moisture content (MC₀) were considered as inputs, while the moisture content output from the dryer (MC_f) was considered as the model's output.

Results and Discussions

The Fig. (9-a), (9-b), and (9-c) show images captured by the first, second, and third cameras, respectively, for under-drying conditions (conveyor speed of 1.7 cm/min), desired drying conditions (conveyor speed of 0.6 cm/min), and over-drying conditions (conveyor speed of 0.35 cm/min) at a temperature of 50 °C. The surface shrinkage of the samples is clearly visible from the images. The images captured by the first camera have similar conditions in almost all experimental treatments because the initial samples entering the dryer were identical. However, variations in the appearance of the samples are observed in the images captured by the second and third cameras. These variations are due to the loss of moisture, which is manifested in the form of shrinkage of the samples.

Table 3 presents the results of training the model for predicting moisture content. By examining the results obtained from this modeling, the network with a structure of 4-12-1 was found to be the best model with RMSE values of 0.0657, 0.2494, and 0.4657, and R values of 0.9999, 0.9999, and 0.9998 for training, evaluation, and testing, respectively.