Early nondestructive testing method for bergamot pears with internal diseases based on vibration signal feature image

Abstract

The invention discloses an early nondestructive testing method and device for bergamot pears with internal diseases based on feature images of vibration signals, and the method comprises the steps: collecting vibration signals of mildewed pears and healthy pears, and converting the vibration signals into corresponding feature images through an image coding method; and then taking the feature image as input to construct a deep learning internal disease bergamot pear discrimination model so as to realize accurate discrimination of early mildewed pears. The method has excellent performance in pear fruit quality classification and prediction, improves the commodity rate and market competitiveness of the pear fruits, and has great significance in promoting rapid and healthy development of the pear fruit industry.

Description

Technical Field

[0001] The present invention belongs to the technical field of rapid non-destructive detection of agricultural product quality, and in particular relates to a non-destructive detection method and device for early internal diseases of pear fruit based on vibration signal characteristic images. Background Art

[0002] The Korla fragrant pear (Pyrus bretschneideri Rehd) is beloved by consumers worldwide for its rich nutritional profile and delicious flavor. However, during growth and storage, the pear is susceptible to pathogenic infection, leading to postharvest diseases that increase the risk of quality loss and reduced exportability. Internal disease is one of the most serious pathological diseases of pear fruit. Internal disease is a typical endogenous disease that develops initially in the kernel and then erodes the flesh outward. Pears infected with internal disease exhibit no obvious external symptoms in their early stages, making them difficult to distinguish from healthy pears. Because internally diseased pears are highly contagious, they can further rot due to the growth of fungi and bacteria, eventually infecting an entire batch of healthy pears and causing significant economic losses. Notably, when internally diseased pears are mixed with healthy pears during further processing, the moldy tissue, containing multiple toxins, can exceed the regulatory limits for mycotoxins in derivative products such as concentrated pear juice and pear wine, posing a potential food safety hazard. Traditionally, internal pear disease detection involves destructive sampling combined with visual inspection by trained personnel, followed by overall assessment. However, these methods are time-consuming, labor-intensive, inefficient, and subjective. Therefore, it is of practical significance to develop a rapid, non-destructive detection technology to detect moldy pear cores early, thereby removing diseased fruit from the post-harvest storage and distribution chain.

[0003] With the emergence of emerging technologies, numerous studies have utilized X-ray technology, near-infrared spectroscopy, hyperspectral imaging, and electronic noses for nondestructive detection of internal fruit diseases. Compared to other nondestructive testing techniques, vibration detection is increasingly being used to assess fruit quality attributes due to its simplicity, rapidity, and low cost. However, research on the use of vibration detection in identifying internally diseased fragrant pears is limited. Notably, several researchers have demonstrated the feasibility of using vibration detection to identify internal fruit diseases. Zhang et al. (2021) applied vibration detection to identify mild browning in pears, achieving a classification accuracy of 86.4%. Han et al. (2023) used this method to detect moldy kernels in shelled pecans, achieving an accuracy of 91.67%. Most of these reported studies focused on using traditional machine learning algorithms. The effectiveness of these algorithms relies heavily on the extraction and selection of features that reflect disease information. These manually extracted features require domain expertise and experience, significantly impacting model performance and generalization. Feature images, as two-dimensional data structures, can capture richer and more complex information features, including spatial distribution, color information, and texture characteristics. If the one-dimensional vibration signal is converted into a feature image and combined with a deep learning model, we can have a more comprehensive understanding of the development degree of internal diseases in pear fruits, which will help to significantly improve the detection accuracy of early internal diseases of pear fruits. This is a research idea worth exploring for the identification of fragrant pears with early internal diseases. Summary of the Invention

[0004] The main technical problem solved by the present invention is to provide a method and device for non-destructive detection of early internal diseases of pear fruits based on characteristic images of vibration signals. The one-dimensional vibration signal is converted into a characteristic image, and an early discrimination model for fragrant pears with internal diseases is constructed. This lays a solid technical foundation for the research and development of early online detection devices for fragrant pears with internal diseases, and solves the current problem of low accuracy in early discrimination of fragrant pears with internal diseases.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows: comprising the following steps: Step 1: Korla fragrant pears were used as the research objects. They were picked at Xinyang Family Farm. The farmers selected pears suspected of internal diseases and healthy pears, and then stored them in a fruit fresh-keeping warehouse at -2~0℃ and relative humidity of 85%~95%. Step 2: Under stable and reliable experimental conditions, using the established vibration response detection system, a pear fruit sample was first placed on the surface of a vibration exciter (YE-5, Beijing Yiyang Yingzhen Testing Technology Co., Ltd., China). An accelerometer (YA19S, Beijing Yiyang Yingzhen Testing Technology Co., Ltd., China) was adjusted to contact the surface of the apple sample. A signal generator (UTG 2000A, Beijing Yiyang Yingzhen Testing Technology Co., Ltd., China) was then activated to generate a raw excitation signal. This raw excitation signal was linearly amplified by a power amplifier (GF-100, Beijing Yiyang Yingzhen Testing Technology Co., Ltd., China) and then transmitted to the vibration exciter. The pear fruit sample then vibrated in response to the vibration of the exciter. Once the signal stabilized, a dynamic signal acquisition card (YSV, Beijing Yiyang Yingzhen Testing Technology Co., Ltd., China) was used to acquire the sample's response signal. Finally, vibration signal analysis software was used to collect the signal. Step 3: Use image coding methods, recursive graph (RP), Gram angle field (GAF) and Markov transition field (MTF) to convert the obtained pear vibration signal into the corresponding feature image; Step 4: Cut the pear in half along the equator and take a photo of the cross section with a digital camera. Use the Matlab 2023b image processing toolbox to extract the RGB three-channel components of the color image, then use the iterative threshold segmentation method to segment the B channel image, and use the 8-connected region labeling algorithm to extract the mold core area in the B channel image. S 1 and the cross-sectional area of ​​the pear S The ratio of 2 is defined as the degree of pear core mold S Step 5: Use image coding method, recursive graph, Gram angle field and Markov transition field to convert the collected vibration signal into feature image to establish vibration signal feature image dataset; Step 6: Based on the established feature image dataset, a deep learning algorithm is used to construct a pear fruit internal disease discrimination analysis model; Step 7: After determining the optimal model for early detection of internal diseased pears, optimize the model using an optimization algorithm.

[0006] The image encoding methods used to convert the vibration signal into a feature image in step three are: recurrence plot (RP), Gram Angular Field (GAF), and Markov Transition Field (MTF). A recurrence plot is a visualization tool used to analyze time series signals, primarily revealing their dynamic characteristics by reconstructing the signal's state space. It maps the time series into a multidimensional space, displaying the signal's phase relationships and periodicity. The points in the recurrence plot represent the signal state at different time points, while the connecting lines reflect the signal's changes over time. Recurrence plots are particularly suitable for capturing nonlinear characteristics in complex signals and can reveal periodic patterns and trends hidden within them. When processing non-stationary signals, recurrence plots can effectively demonstrate the system's dynamic behavior, helping researchers identify the system's stability and chaotic characteristics. By analyzing recurrence plots, we can gain a deeper understanding of the system's interactions and evolution, providing an important basis for further signal processing and prediction. The Gram Angular Field is a time-frequency analysis method based on the Gram matrix. It reveals signal similarities and changing trends by calculating the angle between the signal's eigenvectors at different time points. The Gram angle field effectively captures both the local characteristics and global structure of a signal, making it particularly suitable for analyzing nonstationary and complex signals. This method helps identify periodicity and local patterns in the signal, facilitating subsequent feature extraction and classification. The Markov transition field is a time series signal analysis method based on the Markov process. It constructs a Markov transition matrix to describe the transition probabilities between system states. It can capture the dynamic characteristics of a signal and reveal the evolution of the system state. Markov transition fields are widely used in time series analysis, suitable for predicting future states and identifying system stability and chaotic behavior. They provide an effective methodology for modeling and controlling complex systems.

[0007] The construction strategy for the pear internal disease discrimination analysis model based on the vibration feature image set in step 6 is as follows: ① Using the Random Split algorithm, the 375 fragrant pear vibration multi-domain dataset is randomly divided into a training set and a test set in an 8:2 ratio. ② The vibration signals are converted into vibration feature images using recursive graphs, Gram angle fields, and Markov transfer fields. ③ Using the vibration feature images as input, ResNet, DenseNet, and SqueezeNet models are constructed for early detection of fragrant pear internal diseases. ④ The optimal input feature parameters are determined based on the accuracy of the training set. The deep learning model most suitable for early detection and classification of fragrant pear internal diseases is selected using the accuracy, recall, and F1 model performance evaluation metrics.

[0008] The optimization strategy for the optimal internally diseased fragrant pear early detection model in step 7 is as follows: ① Replace the traditional convolution in the deep learning model with depthwise separable convolution (DSC) to further improve the performance of the classification model for early detection of internally diseased fragrant pears. ② Evaluate the improved classification model based on the optimization algorithm using precision, recall, and F1 model performance evaluation metrics to determine the final deep learning model for early detection and grading of internally diseased fragrant pears.

[0009] The beneficial effects of the present invention are:

[0010] The present invention utilizes recursive graph, Gram angle field and Markov transition field to convert vibration signal into characteristic image and establishes a discrimination model with vibration characteristic image as input.

[0011] This method is used for early nondestructive detection of pears with internal diseases. Because early internal disease symptoms are subtle and difficult to identify with the naked eye, their presence among healthy pears not only affects the overall quality of the pears and negatively impacts the reputation of fruit merchants, but also makes them susceptible to rotting during storage, leading to significant economic losses for merchants. Therefore, using this method for nondestructive detection of pears with internal diseases can effectively eliminate them in the early stages. This paper applies deep learning algorithms in combination with image coding technology to the construction of an early detection model for fragrant pears with internal diseases, fully leveraging the advantages of deep learning in processing two-dimensional images, improving the accuracy of detection, and providing a research idea worthy of exploration for the early identification of fragrant pears with internal diseases. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 It is a flow chart of the feature image conversion method of the present invention. DETAILED DESCRIPTION The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more precise definition of the protection scope of the present invention.

Claims

1. A nondestructive detection method for internal diseased pears based on vibration signal characteristic images, characterized in that: Follow these steps: Step 1: Korla fragrant pears were harvested from a family farm in Awati, Xinjiang. Farmers selected pears suspected of internal disease and healthy pears, which were then stored in a fruit storage facility at -2-0°C and a relative humidity of 85%-95%. Step 2: Under stable and reliable experimental conditions, using the established vibration response detection system, a pear fruit sample was first placed on the surface of a vibration exciter (YE-5, Beijing Yiyang Yingzhen Testing Technology Co., Ltd., China). An accelerometer (YA19S, Beijing Yiyang Yingzhen Testing Technology Co., Ltd., China) was adjusted to contact the surface of the apple sample. A signal generator (UTG2000A, Beijing Yiyang Yingzhen Testing Technology Co., Ltd., China) was then activated to generate a raw excitation signal. This raw excitation signal was linearly amplified by a power amplifier (GF-100, Beijing Yiyang Yingzhen Testing Technology Co., Ltd., China) and then transmitted to the vibration exciter. The pear fruit sample then vibrated in response to the vibration of the exciter. Once the signal stabilized, a dynamic signal acquisition card (YSV, Beijing Yiyang Yingzhen Testing Technology Co., Ltd., China) was used to acquire the sample's response signal. Finally, vibration signal analysis software was used to collect the signal. Step 3: Using image coding methods, recursive graph (RP), Gram angle field (GAF), and Markov transition field (MTF), the obtained pear vibration signal is converted into the corresponding feature image; Step 4: Cut the pear in half along the equator and take a photo of the cross section with a digital camera. Use the Matlab2023b image processing toolbox to extract the RGB three-channel components of the color image, then use the iterative threshold segmentation method to segment the B channel image, and use the 8-connected region labeling algorithm to extract the mold core area in the B channel image. S 1 and the cross-sectional area of ​​the pear S The ratio of 2 is defined as the degree of pear core mold S Step 5: Convert the vibration signal to obtain vibration feature images of three categories of internally diseased fragrant pears, so as to establish a vibration feature image dataset of internally diseased fragrant pears; Step 6: Use the established vibration feature image dataset as input to build a deep learning model to identify pears with internal diseases; Step 7: After determining the optimal model for early detection of internal diseased pears, optimize the model using an optimization algorithm.

2. The method and device for nondestructive detection of early internal diseases of fragrant pears based on vibration signal characteristic images according to claim 1, characterized in that: Repeat step 2 to collect vibration signals from the pears according to the requirement that the ratio of healthy pears, pears with mild internal diseases, and pears with moderate to severe diseases is 1.6:1.4:1; The method and device for early nondestructive detection of internal diseased fragrant pears based on vibration signal characteristic images according to claim 1 are characterized by: In the step 5, based on the completion of the above-mentioned vibration signal conversion, a vibration signal feature image dataset is constructed; The method and device for early nondestructive detection of internal diseased fragrant pears based on vibration signal characteristic images according to claim 1 are characterized by: The deep learning algorithms in step 6 include residual neural network (ResNet), dense connection network (DenseNet), and squeeze neural network (SqueezeNet) deep learning algorithms; The method and device for early nondestructive detection of internal diseases of fragrant pears based on vibration signal characteristic images according to claim 1 are characterized by: The model optimization algorithm in step seven includes replacing traditional convolution with depthwise separable convolution (DSC).

3. A method and device for implementing the vibration signal characteristic graph-based early nondestructive detection of pears with internal diseases as described in claim 1, wherein the early detection system for pears with internal diseases comprises a vibration signal system, a characteristic image conversion system, and a pattern recognition module; wherein the acoustic signal acquisition device is a laser Doppler vibrometer detection device; the characteristic image conversion system converts the vibration signal into a characteristic image using recursive graph (RP), Gram angle field (GAF), and Markov transition field (MTF) algorithms; deep learning models ResNet, DenseNet, and SqueezeNet are constructed for early discrimination of pears with internal diseases, and finally a computer determines the disease extent of the pears being tested.

Images