A new edge detection operator based on dual parity morphological gradients is proposed, in order to accurately segment a porcelain insulator string in an infrared image before the identification and diagnosis of deteriorated insulators by the extracted thermal information. When applied to the double-shed insulators, the proposed detector is capable of effectively eliminating the blurred region between the upper and lower sheds as well as preventing over-segmentation by combining both odd and even structural elements. Compared with the multiscale morphological gradient algorithm, this technique reduces the size of the maximum structural elements to optimize the computational complexity. A quantitative evaluation using PSNR, IoU, and FPR indexes shows that the new technique outperforms other common edge detectors. In addition, the experiments on thousands of real infrared images demonstrate the detector to be highly effective despite its low complexity.

To realize an embedded online diagnosis system for accurate detection of internal defects in high voltage polymeric insulators, a novel ultra-wideband (UWB) microwave diagnostic system for fault detection is proposed in this paper. The UWB system is based on the multimode transferring theory on dielectric waveguides, where the level difference of transmission signals of mode transferring has been employed to detect internal defects, such as cracks. First, an investigation by simulations is performed regarding the interfacial millimeter-order cracks in both transverse and longitudinal orientations. Differences of 30∼50 dB and 20∼40 dB in receiving signals of mode-transferring between non-crack and crack cases have been observed over 3-10 GHz to transverse and longitudinal oriented cracks, respectively. Then, a test prototype is designed and manufactured for experimental verification. The measurement results have a significant 40∼60dB receiving signal level difference between the non-crack reference case and the crack case. The proposed method is promising for further development of online systems to detect millimeter-order cracks in energized polymeric insulators.

Moisture content is an important parameter of plant and matters in agriculture and forestry. In this article microwave technology is used for the measurement of leaf moisture content. A moisture measurement system based on microwave pseudo waveguide is proposed. The system includes transmission antenna, pseudo wave guide structure and receiver antenna. Scattering parameters with sweeping frequency are utilized to characterize the interaction between the microwave electromagnetic field and the leaves under test. The acquired microwave signals are then optimized with frequency band selection method. The data derived from the microwave method are compared with the standard gravitic method. The Long Short-Term Memory (LSTM) neural network algorithm is used to establish the moisture content prediction model, and displays good predictive capability with coefficient of determination (R ² = 0.9178), root mean square error (RMSE = 0.0343), mean absolute error (MAE = 0.0271). The data measured by the microwave method agree well with those by the standard gravitic method. The microwave pseudo waveguide sensing together with the data processing method is nondestructive and fast, which is promising for measurement of the moisture content of plant leaves in agriculture, forestry and food industry.

Ethanol concentration measurement is critical for many industrial applications. This article presents a novel bonnet probe microwave sensor which has a simple, compact dual-probe stainless steel structure. The designed sensor has specific dimensions with an outer diameter of 50 mm and a height of 110 mm. The system operates by extracting multi-frequency scattering parameters (S₁₁ and S₂₁) across a broad frequency range of 0.3–6 GHz. To accurately model the nonlinear relationship between microwave signals and concentration, three deep learning algorithms—Long Short-Term Memory (LSTM), Autoencoder, and Gated Recurrent Unit (GRU)—were employed separately. Experimental results demonstrate that the sensor exhibits high sensitivity, particularly within the optimized 2–6 GHz frequency band. Among the models, the LSTM network achieved the best performance with a coefficient of determination (R²) of 0.9976, a root mean squared error (RMSE) of 0.0134, and a mean absolute error (MAE) of 0.0099. These metrics validate the sensor's advantages in terms of high precision, electromagnetic stability, and suitability for real-time detection in industrial settings.