An intelligent tactile-sensing system can be expected to detect the fruits in modern agriculture. However, conventional optical non-destructive testing cannot fully meet the large-scale production requirements in recent years. The near-infrared spectroscopy has been constrained by fruit surface wax thickness and light scattering artifacts. Machine vision is also limited to the superficial morphological features. Therefore, the critical deficiencies can be found to characterize the internal structural evolution and resolve the sub-Newton firmness gradients. In this study, a tri-finger polyurethane Fin Ray flexible gripper can be expected to overcome some limitations. The Amor-SE-CNN framework was also proposed to assess the fruit quality. Multiresolution time-frequency analysis was converged with adaptive attention mechanisms. A vibration-dynamics approach was established to classify the precision maturity. The optical variables were reduced to maintain the non-destructive integrity. The hardware architecture was integrated with the strain gauges (1.2 cm×1.0 cm sensing area) epoxy-encapsulated at 4.62 cm from the gripper fingertips—an optimal position after the finite-element simulations, indicating the maximum deformation amplitude. During step-motor-controlled grasping sequences (0–12 mm/s closure velocity regulated by DM422 driver, 15 mm stroke), triaxial strain signals were recorded to determine as the four-stage preprocessing: (1) Transient artifact removal via slope-threshold interpolation; (2) fourth-order bidirectional Butterworth bandpass filtering (0.5–5.0 Hz) suppressing >5.0 Hz mechanical vibrations and <0.5Hz thermal drift; (3) Hilbert-transform envelope extraction isolating viscoelastic relaxation; and (4) amplitude normalization dynamically mapped to [0,1] range using piecewise linear scaling. Continuous wavelet transform (CWT) with the complex Morlet wavelets was used to transform 1D strain data into 224×224 pixel time-frequency matrices using logarithmic energy spectrum computation and bilinear interpolation. Three-channel RGB space fusion was performed on the spectrograms. The channel-specific energy distributions were encoded within the biomechanically critical 0.5–5.0 Hz band into composite color-textural signatures. There were the stiffness-dependent frequency modulations—exemplified by overripe fruits with the 0.5–1.5 Hz dominant energy versus hard-unripe specimens concentrating at 2.5–5.0 Hz. The convolutional neural network was employed as a squeeze-and-excitation attention module. The global context aggregation (GAP→8D descriptor→sigmoid-activated 32D reconstruction) was implemented to adaptively amplify the firmness correlated spectral components. While 3×3 dynamic convolution kernels with ReLU activation enhanced the spatial sensitivity to localized energy discontinuities. Training was incorporated to enhance the multi-strategy robustness: Stochastic data augmentation (±10% random cropping, ±20% brightness jitter, ±15% contrast modulation) was simulated for field operation variances; 50% Dropout regularization was countered the small-sample overfitting; and Adam optimization was used to minimize the categorical cross-entropy across 100 epochs with early stopping. The validation was involved 420 kiwifruits ('Yangtao Bao': n=240; 'Hayward': n=180) with five physiological maturity tiers (F<9.4N: overripe; 9.4N≤F<11.3N: ripe; 11.3N≤F<13.7N: mid-ripe; 13.7N≤F<16.7N: unripe; F≥16.7N: hard-unripe), according to the GY 4 texture analyzer reference measurements. The results show that the Amor-SE-CNN achieved a 93.3% classification accuracy—surpassing the conventional CNN (84.8%), SE-CNN (88.6%), and time-frequency CNN (90.5%) baselines by 8.5, 4.7, and 2.8 percentage points, respectively, while outperforming prior tactile studies. Attention mechanisms were specifically enhanced the discrimination between the transitional maturity states. The "soft" vs "mid-ripe" F1 scores were elevated from 81% to 92% through 3–4 Hz band amplification. Physiological integrity was confirmed via respiration kinetics: CO₂ evolution rates shared no statistically significant intergroup variance (P＞0.05) during 72 h monitoring, thus verifying negligible mechanical stress. An experimental platform was constructed to detect the fruit firmness using a flexible gripper. Time-frequency analysis was integrated with an attention-enhanced Convolutional Neural Network (CNN). The effective classification of kiwifruit maturity was achieved after enhancement. The finding can also provide technical support for the intelligent post-harvesting of the fruits.

Citrus Huanglongbing (HLB) is one of the most serious plant diseases in modern agriculture. Hyperspectral reflectance can contain the physical morphological features of the leaf surfaces. This study aimed to improve the accuracy of the non-destructive diagnosis for the Citrus HLB. Hyperspectral reflectance was also fused with the transmittance information of the sample surface, particularly for the internal structural and compositional features. A hyperspectral reflectance-transmittance fusion was proposed to detect the HLB in the citrus leaves. Firstly, the polymerase chain reaction (PCR) was used to identify the severity levels of the citrus HLB in leaves. After that, the leaves were categorized into four classes: healthy, mild infection, moderate infection, and severe infection. Subsequently, the reflectance and transmittance hyperspectral data in the wavelength range of 372.66-1 039.65 nm of the citrus leaves were collected at the different severity levels of the citrus HLB. The hyperspectral data were preprocessed using standard normal variate (SNV), savitzk-golay (SG) filtering, and multiplicative scatter correction (MSC). Then, the uninformative variable elimination (UVE), competitive adaptive reweighted sampling (CARS), and successive projections algorithm (SPA) were adopted to conduct the feature variable selection for the preprocessed data. Three machine learning methods—least squares support vector machine (LS-SVM), random forest (RF), and back-propagation neural network (BPNN)—were utilized to perform the classification modeling. Meanwhile, the spectral analysis was conducted on the single reflectance and transmittance hyperspectral data for the leaves at the various HLB levels. The classification models were established using reflectance and transmittance spectra. Finally, three grading models were developed for the citrus HLB severity after hyperspectral reflectance-transmittance fusion (data-, feature-, and decision-level fusion). The experimental results showed that the single reflectance spectra, single transmittance spectra, and reflectance-transmittance fused classification were all effectively classified and then detected the citrus HLB. The feature and decision-level fusion significantly outperformed the single-spectral approaches. Particularly, the decision-level fusion achieved the best prediction performance at the different severity levels of the HLB under various models. Furthermore, the prediction accuracy was also improved by 3.2% to 11.6%, compared with the single-spectrum data. Among them, the decision-level fusion with the BPNN classification model—where SG filter preprocessing was combined with the CARS for the characteristic band selection—was achieved in the best classification, with an accuracy rate of 99.98%. Hyperspectral imaging technology with the reflectance-transmittance fusion can be expected for the effective, rapid, and high-precision detection of the citrus HLB severity. Decision-level fusion can maximize the performance of the model. The findings can also provide a strong reference to detect plant diseases.

The size and shape of individual fruit cells are key indicators of a fruit’s physiological condition and overall quality. However, due to the three-dimensional (3D) nature of fruit cells, existing biomicroscopes are not capable of efficiently or accurately characterizing their 3D geometry. In this work, a novel microscope system integrated with computer software was developed to enable precise 3D geometrical characterization of fruit cells. To validate the system’s effectiveness, tomatoes and strawberries at two different stages of ripeness were used as test samples. First, the front and bottom views of the fruit cells were captured. Subsequently, the developed software was used to measure the 3D geometric size of each individual cell. Key performance parameters of the developed 3D microscope, including overall magnification, aperture diameter, resolution, and field of view area, were carefully measured and evaluated. The experiment revealed differences in the length D₁, thickness D₂, width D₃, and geometric mean diameter (GMD) of single cells of tomatoes and strawberries; these differences were 18.18%, 4.6%, 9.8%, and 29.7%, 10.7%, 12.6%, respectively. Furthermore, 3D geometrical data, including surface area S, volume V, and sphericity φ of single cells, were successfully obtained. This demonstrated that the developed microscope system can efficiently and accurately capture and characterize the true 3D geometry of cells, emphasizing its scientific value.

Mechanical bruising has been one of the main influencing factors of yellow peaches on consumer satisfaction and shelf life. The implicit bruises can often be suffered during picking, grading, and transportation. Early slight bruising can accelerate the spoilage and decay under the action of biological enzymes, due to the fungal infection. Therefore, it is crucial to timely and efficiently detect the implicit bruises in yellow peaches. However, the manual identification of surface bruises on fruits and vegetables cannot fully meet the large-scale production at present, due to the inefficiency, cost, and easy influence by subjective human factors. Common non-destructive testing techniques also share some limitations in detecting fruit and vegetable bruises, such as machine vision, multispectral/hyperspectral, X-ray/CT, thermal and fluorescence imaging. Structured illumination reflectance imaging (SIRI) is characterized by a non-contact, wide field of view and depth discrimination. The spatial frequency of structured light and the penetration depth of light in biological tissues can be regulated to overcome the limitation of conventional uniform illumination imaging that cannot obtain depth information. The SIRI can be combined the image processing and classification, in order to effectively detect the implicit defects of the fruits and vegetables. However, it is usually required to obtain three-phase shifted structured illumination images, which is limited to the imaging and detection speed. In this study, the SIRI technology was combined with the SPT fast demodulation and machine learning, and then rapidly detected the early implicit bruises in yellow peaches. The feasibility of hidden damage detection was verified using SIRI technology and two SPT demodulations. The detection efficiency of SIRI technology was improved for the online detection of implicit bruises in yellow peaches. Firstly, the SIRI system was used to collect the stripe-structured light reflection images at the spatial frequencies of 0.05, 0.10, 0.15, 0.20, 0.25, and 0.30 mm⁻¹. Three-phase demodulation (TPD) was used to obtain the amplitude component (AC) and direct component (DC) images. The contrast index (CI) of the AC image was calculated to select the optimal spatial frequency. Three-phase stripe images were collected on all samples. The AC and DC images were obtained after capture. The ratio image (RT) of AC/DC was calculated to improve the uneven surface brightness of yellow peaches after SPT demodulation. According to the gray level co-occurrence matrix (GLCM) of DC, AC, and RT images, the local binary pattern (LBP) image texture features, and depth features extracted by ResNet-50, the inputs were taken as the GLCM-LBP, depth and mixed features of five images (DC, AC, RT, DC-AC, and DC-AC-RT). The healthy and bruised yellow peaches were classified using machine learning models, such as the support vector machine (SVM), K-Nearest Neighbor (KNN), XGBoost, and random forest (RF). The results showed that the highest average accuracies of the improved model were 92.6%, 95.0%, and 95.7%, respectively, for the detection of bruised peaches using GLCM-LBP, depth, and mixed features. The mixed feature model achieved the highest average accuracy. Therefore, the GMLC-LBP and depth features were combined to effectively improve the classification accuracy of the model. The XGBoost model with the DC-AC-RT image features also shared the highest accuracy of 97.6%. Furthermore, the overall classification accuracy of the SPT demodulated images was comparable to that of TPD ones, with the highest accuracy of 97.6%. The image acquisition time was also saved by one-third, with only any two-phase shifted images. In summary, the SIRI with the SPT and machine learning was achieved to detect the implicit bruises in peaches, with high accuracy and less detection time, thus improving the detection efficiency of SIRI technology. This finding can also provide a strong reference for the real-time detection of implicit bruises on fruits and vegetables.