In-situ non-destructive method for measuring fruit firmness on a tree

Abstract

The application discloses a kind of in-situ nondestructive testing methods of fruit hardness on tree, which can be used to detect the hardness of apple, honey peach, kiwi, plum, mango, citrus and pear and other fruits on tree.A non-contact fruit vibration response detection device on tree is constructed to measure the vibration response signal of fruit on tree in-situ.The wavelet threshold denoising method is used to denoise the fruit vibration response on tree, and the autoregressive model method is used to calculate the power spectral density of denoised fruit vibration response.A one-dimensional convolutional neural network model with multi-scale receptive field is constructed, which is suitable for locating the multi-scale deep features related to fruit hardness in power spectral density under the condition of variable excitation distance and environmental wind interference leading to the uncertainty of fruit power spectral density amplitude, and an accurate fruit hardness prediction model on tree is established based on multi-scale feature fusion.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing of fruit quality, and to an in-situ non-destructive testing method for the hardness of fruit on a tree. Background Technology

[0002] Firmness is an important indicator of fruit quality, closely related to its ripeness, taste, and flavor. On-site testing of fruit firmness in the orchard allows for monitoring fruit quality and guiding refined management of individual trees. It also provides data support for implementing staggered harvesting based on diversified sales strategies, achieving an optimal balance between storage and transportability and flavor. Furthermore, it assesses the field performance of varieties, providing data support for the breeding of high-quality fruit varieties.

[0003] In actual orchard production, sampling is often used to harvest fruits from different canopy layers, and portable instruments are used to test fruit firmness. Commonly used portable instruments include handheld puncture meters and acoustic vibration detection devices. Handheld puncture meters assess fruit firmness by measuring the fruit's cracking force; however, this measurement is destructive, and the results are affected by the operator, resulting in low accuracy. Portable acoustic vibration detection devices clamp the fruit between a piezoelectric vibrator and a vibration pickup sensor, extracting characteristic parameters such as amplitude, frequency, and phase from the acquired acoustic vibration response signal based on prior knowledge to determine the fruit's firmness. Although contact vibration sensors can measure the acoustic vibration response of fruit, they need to be directly attached or tightly pressed against the fruit surface, and the sensor's mass can interfere with the fruit's vibration, thus affecting the measurement results. Feature extraction methods based on prior knowledge have weak characterization capabilities for fruit firmness under the influence of complex environmental factors in orchards.

[0004] The purpose of this invention is to overcome the shortcomings in the above-mentioned background technology and provide an in-situ non-destructive testing method for the firmness of fruit on trees. This method allows for non-contact, in-situ, and non-destructive testing of the firmness of fruit on trees without interfering with fruit growth, thus providing technical support for improving the quality and efficiency of the fruit industry. Summary of the Invention

[0005] This invention provides an in-situ non-destructive testing method for the hardness of tree-grown fruits, applicable to various fruits such as apples, peaches, kiwis, plums, mangoes, citrus fruits, and pears. Addressing the technical bottlenecks of lacking non-contact on-site testing devices and effective methods for analyzing the acoustic vibration response of non-stationary fruits and constructing fruit hardness models, this invention constructs a non-contact acoustic vibration response testing device for in-situ measurement of the acoustic vibration response signal of tree-grown fruits. A wavelet threshold denoising method is used to reduce noise in the acoustic vibration response, and an autoregressive model method is employed to calculate the power spectral density of the denoised acoustic vibration response. A novel one-dimensional convolutional neural network (CNN) model with multi-scale receptive fields is constructed. mUsing power spectral density as input to predict fruit firmness, this method is suitable for situations where the amplitude of power spectral density of fruit on the tree is uncertain due to changes in excitation distance and environmental wind interference. It can locate multi-scale depth features related to fruit firmness in the power spectral density and establish an accurate prediction model for fruit firmness on the tree based on the fusion of multi-scale features.

[0006] A non-destructive in-situ testing method for the firmness of tree-grown fruit, characterized by comprising the following steps:

[0007] (1) A non-contact acoustic vibration response detection device for tree fruits was constructed, consisting of a jet excitation unit and a vibration measurement unit. The jet excitation unit generates instantaneous pressurized airflow to excite the tree fruits, while the non-contact vibration measurement unit measures the acoustic vibration response of the fruits.

[0008] The fruits on the tree are apples, peaches, kiwis, plums, mangoes, citrus fruits, or pears.

[0009] (2) The wavelet threshold denoising method was used to reduce the noise of the fruit acoustic response, and the denoised acoustic response of the fruit on the tree was obtained.

[0010] The jet excitation unit includes:

[0011] An air compressor used to generate pressurized airflow;

[0012] A pressure regulating valve is installed on the air compressor;

[0013] A solenoid valve for controlling the gas flow is connected to the pressure regulating valve via a pipeline.

[0014] The nozzle is connected to the solenoid valve via a pipeline;

[0015] A laser rangefinder sensor is used to measure the distance between the nozzle and the fruit on the tree.

[0016] The vibration measurement unit includes:

[0017] A laser Doppler vibration meter for measuring the acoustic vibration response of fruit on trees;

[0018] The controller connected to the laser Doppler vibration meter;

[0019] The data acquisition card is connected to the controller;

[0020] A computer connected to the data acquisition card.

[0021] (3) Using the autoregressive model method with statistical characteristics, calculate the power spectral density of the acoustic vibration response of the tree fruit after noise reduction obtained in step (2), and analyze the frequency domain characteristics of the acoustic vibration response.

[0022] (4) Construct a novel one-dimensional convolutional neural network model with multi-scale receptive fields. Input the power spectral density of the denoised acoustic vibration response of the tree fruit obtained in step (3) into the one-dimensional convolutional neural network model with multi-scale receptive fields to obtain the hardness of the tree fruit.

[0023] In step (2), wavelet threshold denoising is used to reduce the noise in the fruit's acoustic response, resulting in the denoised acoustic response of the fruit on the tree. Specifically, this includes:

[0024] The fruit sound vibration is decomposed into n-1 orders using the db6 wavelet function, yielding n wavelet coefficients. Let s be the sum of squares of the n wavelet coefficients, and let... , , and The Heursure threshold decision coefficient is... Then the Heursure threshold is calculated as follows: , The standard deviation of noise at the decomposition scale; if Then the squares of the wavelet coefficients w1, w2, … …, w n Sort by size in ascending order and calculate the risk vector Q(i):

[0025]

[0026] Find the value that minimizes the risk vector Q(i). Element position k min And calculate the Heursure threshold as ;

[0027] Subtract the Heursure threshold from the wavelet coefficients that are greater than or equal to the Heursure threshold, and set the wavelet coefficients that are less than the Heursure threshold to 0 to obtain the denoised wavelet coefficients. Reconstruct the denoised wavelet coefficients to obtain the denoised fruit acoustic response.

[0028] The Heursure threshold is calculated based on two methods to determine the optimal threshold, when... At that time, the threshold is calculated based on the position corresponding to the minimum risk vector. When the noise level is high, a fixed threshold is calculated. The threshold calculation method described above is adaptive; different thresholds will be calculated for fruit acoustic responses with varying noise levels. Subtracting the Heursure threshold from the wavelet coefficients greater than or equal to it, and setting wavelet coefficients less than the Heursure threshold to 0, results in wavelet coefficients with better continuity and smoother signal reconstruction. Therefore, the wavelet thresholding denoising method can effectively remove noise from the fruit acoustic response.

[0029] Further optimization yields n-1 of 5 to 9. The optimal value is n-1 of 7. The 8 wavelet coefficients include one low-frequency coefficient cA1 and seven high-frequency coefficients cD1, cD2, cD3, cD4, cD5, cD6, and cD7.

[0030] In step (3), the power spectral density of the fruit's acoustic vibration response after noise reduction is calculated using a 15th-order autoregressive model.

[0031] In step (4), the one-dimensional convolutional neural network model with multi-scale receptive fields specifically includes:

[0032] An input layer;

[0033] The first convolutional layer connected to the input layer;

[0034] The first pooling layer connected to the first convolutional layer;

[0035] A second convolutional layer connected to the first pooling layer, the second convolutional layer comprising 4 convolutional kernels;

[0036] A third convolutional layer connected to the second convolutional layer;

[0037] A splicing layer connected to the second and third convolutional layers;

[0038] The second pooling layer is connected to the splicing layer;

[0039] The discard layer connected to the second pooling layer;

[0040] The flattened layer connected to the discard layer;

[0041] The first fully connected layer connected to the flattened layer;

[0042] A second fully connected layer connected to the first fully connected layer;

[0043] The output layer connected to the second fully connected layer.

[0044] The one-dimensional convolutional neural network model with multi-scale receptive fields of this invention is suitable for locating multi-scale deep features related to fruit hardness in the power spectral density when the amplitude of the power spectral density of fruit on the tree is uncertain due to changes in excitation distance and environmental wind interference. Based on the fusion of multi-scale features, an accurate prediction model for the hardness of fruit on the tree is established.

[0045] Further optimization reveals an in-situ non-destructive testing method for the firmness of tree-grown fruit, comprising the following steps:

[0046] (1) Constructing a non-contact acoustic vibration response detection device for tree fruits. The acoustic vibration response detection device for tree fruits consists of two parts: a non-contact jet excitation unit and a vibration measurement unit. The jet excitation unit generates instantaneously pressurized airflow to excite the tree fruits, while the non-contact vibration measurement unit measures the acoustic vibration response of the fruits. The jet excitation unit consists of an air compressor, a pressure regulating valve, a solenoid valve, a nozzle, and a laser rangefinder. The air compressor is used to generate pressurized airflow, and a pressure reducing valve is installed at the compressor outlet to ensure the stability of the output airflow pressure. The solenoid valve is used to control the on / off of the output airflow. A stainless steel conical nozzle is used to make the final ejected airflow more concentrated and propagate in a straight line. The laser rangefinder is used to measure the distance between the object being measured and the nozzle to ensure that the excitation distance is within an optimal range to fully excite the tree fruits. The vibration measurement unit consists of a laser Doppler vibrometer, a controller, a data acquisition card, and a computer.

[0047] (2) In order to suppress noise in the acoustic vibration response of tree fruit obtained by the detection device, the present invention adopts wavelet threshold denoising method to reduce noise in the low-frequency and high-frequency coefficients of the acoustic vibration response of tree fruit after wavelet decomposition.

[0048] (3) Since the acoustic vibration response of the fruit on the tree after noise reduction is still non-stationary, it cannot be directly subjected to Fourier transform. This invention uses an autoregressive model method with statistical characteristics to calculate the power spectral density of the acoustic vibration response of the fruit on the tree after noise reduction, thereby analyzing the frequency domain characteristics of the acoustic vibration response.

[0049] (4) Based on the power spectral density of fruit on the tree, this invention constructs a one-dimensional convolutional neural network model (CNNm) with multi-scale receptive fields. It is suitable for locating multi-scale deep features related to fruit hardness in the power spectral density when the amplitude of the power spectral density of fruit on the tree is uncertain due to changes in excitation distance and environmental wind interference. Based on the fusion of multi-scale features, an accurate prediction model for the hardness of fruit on the tree is established.

[0050] The following are preferred technical solutions of the present invention:

[0051] In step (1), when using the constructed device to measure the acoustic vibration response of fruit on the tree in situ, the optimal levels for the three excitation parameters—gas pressure, excitation distance, and excitation time—are within 300 kPa, 200 ms, and 20 cm, respectively. The excitation point and the measurement point are placed on opposite sides of the fruit at a 180° interval. Under these conditions, the signal-to-noise ratio of the acoustic vibration response of the fruit on the tree is the best.

[0052] In step (2), the wavelet threshold denoising method uses the db6 wavelet function to decompose the fruit sound vibration response seven times, obtaining one low-frequency coefficient cA1 and seven high-frequency coefficients cD1, cD2, cD3, cD4, cD5, cD6, cD7. The wavelet coefficients greater than the Heursure threshold are subtracted from the Heursure threshold, and the wavelet coefficients less than the Heursure threshold are set to 0, resulting in the denoised cA1 and cD1, cD2, cD3, cD4, cD5, cD6, cD7. The denoised cA1 and cD1, cD2, cD3, cD4, cD5, cD6, cD7 are reconstructed to obtain the denoised fruit sound vibration response.

[0053] In step (3), based on the Akaike information content criterion, the final prediction error criterion and the minimum description length order criterion, it is determined that the 15th order is the optimal order of the autoregressive model. The three criterion values ​​calculated by the 15th order autoregressive model are the smallest, indicating that the calculated power spectral density has high resolution, smoothness and peak clarity, and low risk of overfitting.

[0054] In step (4), CNNm includes an input layer, a first convolutional layer (Conv1), a first pooling layer, a second convolutional layer (Conv2), a third convolutional layer (Conv3), a concatenation layer, a second pooling layer, a dropout layer, a flattening layer, a first fully connected layer (F1), a second fully connected layer (F2), and an output layer. In Conv2 and Conv3, three parallel convolutional kernels of sizes 1, 3, and 5 are stacked to increase the width of the network. They slide sequentially from left to right and from top to bottom on the power spectral density, performing convolution operations with the convolutional kernels on each covered local matrix and outputting feature maps. This is suitable for situations where the amplitude of the power spectral density of fruit on trees is uncertain due to changes in excitation distance and environmental wind interference. It can locate multi-scale deep features related to fruit hardness in the power spectral density and establish an accurate prediction model for the hardness of fruit on trees based on the fusion of multi-scale features. After each convolutional layer, the ReLU activation function is used to enhance the model's non-linearity, and batch normalization and max pooling layers are added to reduce redundant features, prevent overfitting, and accelerate convergence. After three convolutions, the output feature maps of Conv2 and Conv3 are concatenated through a concatenation layer, and then a batch of neurons is randomly truncated through a dropout layer. Finally, a Flatten layer is used to convert the multidimensional features into a one-dimensional vector, which is then fed into fully connected layers F1 and F2 for weighted feature integration. F1 and F2 contain 50 and 1 neurons, respectively, with F2 outputting the final fruit firmness prediction.

[0055] Compared with the prior art, the present invention has the following advantages:

[0056] This invention provides a method for detecting the firmness of fruit on trees. The measurement does not require contact with the fruit and can accurately detect the firmness of fruit on trees in a non-contact, in-situ, and non-destructive manner without interfering with fruit growth, thus providing technical support for improving the quality and efficiency of the fruit industry. Attached Figure Description

[0057] Figure 1 This is a flowchart of a method for testing the firmness of fruit on a tree.

[0058] Figure 2 This is a schematic diagram of a device for detecting the acoustic and vibration response of fruit on a tree.

[0059] Figure 3 To obtain the power spectrum and signal-to-noise ratio of the acoustic vibration response of fruit on the tree.

[0060] Figure 4 This is the architecture for constructing a one-dimensional convolutional neural network model with multi-scale receptive fields.

[0061] Figure 5 This is a visualization heatmap obtained by weighting the concatenated layers of a one-dimensional convolutional neural network model using the gradient-weighted class activation mapping method.

[0062] In the diagram, 1-air compressor, 2-pressure regulating valve, 3-solenoid valve, 4-nozzle, 5-laser rangefinder sensor, 6-laser Doppler vibration meter, 7-controller, 8-data acquisition card, 9-computer. Detailed Implementation

[0063] The present invention will be further described below with reference to the accompanying drawings, but the present invention is not limited to the following embodiments.

[0064] An in-situ non-destructive testing method for the firmness of tree-grown fruit, such as... Figure 1 As shown, it includes the following steps:

[0065] (1) Construct a non-contact acoustic vibration response detection device for tree-borne fruits. For example... Figure 2As shown, the acoustic vibration response detection device for fruit on a tree consists of two parts: a non-contact jet excitation unit and a vibration measurement unit. The jet excitation unit generates instantaneously pressurized airflow to excite the fruit on the tree, while the non-contact vibration measurement unit measures the acoustic vibration response of the fruit. The jet excitation unit consists of 1. an air compressor, 2. a pressure regulating valve, 3. a solenoid valve, 4. a nozzle, and 5. a laser rangefinder. The air compressor generates pressurized airflow, and a pressure reducing valve is installed at the compressor outlet to ensure the stability of the output airflow pressure. The solenoid valve controls the on / off state of the output airflow. A stainless steel conical nozzle is used to concentrate the final ejected airflow and propagate it in a straight line. The laser rangefinder measures the distance between the object being measured and the nozzle, ensuring that the excitation distance is within an optimal range to fully excite the fruit on the tree. When measuring the acoustic vibration response of the fruit in situ, the excitation point and the vibration measurement point are placed on opposite sides of the fruit at a 180° interval, while the gas pressure is controlled at 300 kPa, the excitation time is controlled at 200 ms, and the excitation distance is controlled within 20 cm. The vibration measurement unit consists of a laser Doppler vibration meter (6), a controller (7), a data acquisition card (8), and a computer (9).

[0066] (2) To suppress noise in the acoustic vibration response of tree fruit obtained using the detection device, this invention employs a wavelet threshold denoising method to reduce the noise in the low-frequency and high-frequency coefficients of the acoustic vibration response of tree fruit after wavelet decomposition. The db6 wavelet, which has a superior denoising effect, is selected to decompose the acoustic vibration response of tree fruit seven times. Wavelet coefficients greater than the Heursure threshold are subtracted from the Heursure threshold, and wavelet coefficients less than the Heursure threshold are set to 0. The denoised wavelet coefficients are then reconstructed to obtain the denoised acoustic vibration response of tree fruit.

[0067] (3) Since the acoustic vibration response of the fruit on the tree after noise reduction is still non-stationary, it cannot be directly subjected to Fourier transform. This invention uses an autoregressive model method with statistical characteristics to calculate the power spectral density of the acoustic vibration response of the fruit on the tree after noise reduction, thereby analyzing the frequency domain characteristics of the acoustic vibration response. The power spectral density of the acoustic vibration response of the fruit after noise reduction is calculated using a 15th-order autoregressive model. Figure 3 The power spectral density of three fruits from trees was displayed. It can be seen that each fruit's power spectral density contains more than three clear resonance peaks, and the frequencies corresponding to each resonance peak can be observed. Furthermore, each fruit exhibits a high signal-to-noise ratio in its acoustic response. As the fruit's hardness increases, the peak value of the power spectral density shifts towards higher frequencies, indicating that the fruit's power spectral density carries information closely related to its hardness. Therefore, the constructed detection device can measure the acoustic response of fruits from trees in situ, the wavelet threshold denoising method can effectively remove noise from the acoustic response, and the autoregressive model method can accurately calculate the power spectral density of the acoustic response.

[0068] (4) This invention constructs a one-dimensional convolutional neural network model (CNNm) with multi-scale receptive fields. The architecture of CNNm is as follows: Figure 4 As shown, the power spectral density of the fruit is used as input to predict the hardness of the fruit on the tree. Figure 5 A visualization heatmap is shown, obtained by weighting the concatenated layers of CNNm using a gradient-weighted class activation mapping method. Darker colors in the graph indicate higher importance for predicting fruit firmness. It can be seen that CNNm can locate important regions related to fruit firmness in the power spectral density, such as peaks and troughs, even when the amplitude of the power spectral density of fruit on the tree is variable due to changes in excitation distance and environmental wind interference. The horizontal axis of the peaks corresponds to the various resonant frequencies of the fruit, indicating that the constructed model can accurately predict the firmness of fruit on the tree.

[0069] This invention uses peaches as an example, measuring the acoustic vibration response of peaches on 285 trees. Wavelet thresholding denoising and an autoregressive model are used to obtain the power spectral density of the acoustic vibration response, generating a structurally sound dataset. The dataset is divided into training and testing sets in a 4:1 ratio for training and testing the model, respectively. This invention uses the root mean square error (RMSE) and coefficient of determination (R²) as the basis for determining the model's performance. 2 The residual prediction bias (RPD) was used to compare the performance of the constructed CNNm model with common machine learning models such as Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), and a single-branch one-dimensional convolutional neural network model (CNNt) in predicting the firmness of fruits on a tree. As shown in Table 1, CNNm achieved the highest prediction accuracy, with the training set... RMSEC and The values ​​are 0.826, 2.263 N / mm, and 2.402, respectively. (CNNm test set) , RMSEP and The values ​​were 0.813, 2.501 N / mm and 2.334, respectively, representing improvements of 9.0%, 14.2% and 16.5% over the best SVR model in machine learning, and improvements of 3.8%, 7.2% and 7.8% over CNNT. The results indicate that the constructed one-dimensional convolutional neural network model CNNm with multi-scale receptive fields is suitable for locating multi-scale deep features related to fruit hardness in the power spectral density under conditions where the amplitude of the power spectral density of fruit on the tree is uncertain due to changes in excitation distance and environmental wind interference. Based on the fusion of multi-scale features, an accurate prediction model for the hardness of fruit on the tree can be established.

[0070] Table 1. Hardness prediction performance of CNNm, SVR, and CNNT models based on the power spectrum of peaches on trees.

[0071]

[0072] The above results demonstrate that the present invention provides an accurate and practical non-contact, in-situ, and non-destructive testing method for the hardness of tree-grown fruits, which can provide important technical support for improving the quality and efficiency of my country's fruit industry.

Claims

1. A non-destructive in-situ method for testing the firmness of tree-grown fruit, characterized in that, Includes the following steps: (1) A non-contact acoustic vibration response detection device for tree fruits was constructed, consisting of a jet excitation unit and a vibration measurement unit. The jet excitation unit generates instantaneous pressurized airflow to excite the tree fruits, while the non-contact vibration measurement unit measures the acoustic vibration response of the fruits. The jet excitation unit includes: An air compressor used to generate pressurized airflow; A pressure regulating valve is installed on the air compressor; A solenoid valve for controlling the gas flow is connected to the pressure regulating valve via a pipeline. The nozzle is connected to the solenoid valve via a pipeline; A laser rangefinder sensor for measuring the distance between the nozzle and the fruit on the tree; (2) The wavelet threshold denoising method was used to reduce the noise of the fruit acoustic response, and the denoised acoustic response of the fruit on the tree was obtained. (3) Using the autoregressive model method with statistical characteristics, calculate the power spectral density of the acoustic vibration response of the tree fruit after noise reduction obtained in step (2), and analyze the frequency domain characteristics of the acoustic vibration response. In step (3), based on the Akaike information content criterion, the final prediction error criterion and the minimum description length order criterion, it is determined that order 15 is the optimal order of the autoregressive model. The power spectral density of the denoised fruit acoustic vibration response is calculated using the order 15 autoregressive model; (4) A one-dimensional convolutional neural network model with multi-scale receptive fields is constructed. The power spectral density of the denoised tree fruit acoustic vibration response obtained in step (3) is input into the one-dimensional convolutional neural network model with multi-scale receptive fields to obtain the hardness of the tree fruit; One-dimensional convolutional neural network models with multi-scale receptive fields specifically include: An input layer; The first convolutional layer connected to the input layer; The first pooling layer connected to the first convolutional layer; A second convolutional layer connected to the first pooling layer, the second convolutional layer comprising 4 convolutional kernels; A third convolutional layer connected to the second convolutional layer; A splicing layer connected to the second and third convolutional layers; The second pooling layer is connected to the splicing layer; The discard layer connected to the second pooling layer; The flattened layer connected to the discard layer; The first fully connected layer connected to the flattened layer; A second fully connected layer connected to the first fully connected layer; The output layer connected to the second fully connected layer.

2. The in-situ non-destructive testing method for the hardness of tree-grown fruit according to claim 1, characterized in that, In step (1), the vibration measurement unit includes: A laser Doppler vibration meter for measuring the acoustic vibration response of fruit on trees; The controller connected to the laser Doppler vibration meter; The data acquisition card is connected to the controller; A computer connected to the data acquisition card.

3. The in-situ non-destructive testing method for the hardness of tree-grown fruit according to claim 1, characterized in that, In step (2), wavelet threshold denoising is used to reduce the noise in the fruit's acoustic response, resulting in the denoised acoustic response of the fruit on the tree. Specifically, this includes: The fruit sound vibration is decomposed into n-1 orders using the db6 wavelet function, yielding n wavelet coefficients. Let s be the sum of squares of the n wavelet coefficients, and let... , , and The Heursure threshold decision coefficient is... Then the Heursure threshold is calculated as follows: , The standard deviation of noise at the decomposition scale; if Then the squares of the wavelet coefficients w1, w2, … …, w n Sort by size in ascending order and calculate the risk vector Q(i): Find the value that minimizes the risk vector Q(i). Element position k min And calculate the Heursure threshold as ; Subtract the Heursure threshold from the wavelet coefficients that are greater than or equal to the Heursure threshold, and set the wavelet coefficients that are less than the Heursure threshold to 0 to obtain the denoised wavelet coefficients. Reconstruct the denoised wavelet coefficients to obtain the denoised fruit acoustic response.

4. The in-situ non-destructive testing method for the hardness of tree-grown fruit according to claim 3, characterized in that, In step (2), n-1 is 5 to 9.

5. The in-situ non-destructive testing method for the firmness of tree-grown fruit according to claim 4, characterized in that, In step (2), n-1 is 7.

6. The in-situ non-destructive testing method for the firmness of tree-grown fruit according to claim 5, characterized in that, In step (2), the 8 wavelet coefficients include 1 low-frequency coefficient cA1 and 7 high-frequency coefficients cD1, cD2, cD3, cD4, cD5, cD6, and cD7.

7. The in-situ non-destructive testing method for the firmness of tree-grown fruit according to claim 1, characterized in that, In step (1), the fruit on the tree is an apple, a peach, a kiwi, a plum, a mango, a citrus fruit, or a pear.

Images