Soil porosity, as an essential indicator for assessing soil quality, plays a key role in guiding agricultural production, so it is beneficial to detect soil porosity. However, the currently available methods do not apply to high-precision and rapid detection of soil with a black-box nature in the field, so this paper proposes a soil porosity detection method based on ultrasound and multi-scale CNN-LSTM. Firstly, a series of ring cutter soil samples with different porosities were prepared manually to simulate soil collected in the field using a ring cutter, followed by ultrasonic signal acquisition of the soil samples. The acquired signals were subjected to three kinds of data augmentation processes to enrich the dataset: adding Gaussian white noise, time shift transformation, and random perturbation. Since the collected ultrasonic signals belong to long-time series data and there are different frequency and sequence features, this study constructs a multi-scale CNN-LSTM deep neural network model using large convolution kernels based on the idea of multi-scale feature extraction, which uses multiple large convolution kernels of different sizes to downsize the collected ultra-long time series data and extract local features in the sequences, and combining the ability of LSTM to capture global and long-term dependent features enhances the feature expression ability of the model. The multi-head self-attention mechanism is added at the end of the model to infer the before-and-after relationship of the sequence data to improve the degradation of the model performance caused by waveform distortion. Finally, the model was trained, validated, and tested using ultrasonic signal data collected from soil samples to demonstrate the accuracy of the detection method. The model has a coefficient of determination of 0.9990 for detecting soil porosity, with a percentage root mean square error of only 0.66%. It outperforms other advanced comparative models, making it very promising for application.

Soil texture is recognised as one of the key physical properties of soil. Although traditional laboratory testing methods can determine soil texture information with high accuracy, they are often considered time-consuming and costly. To achieve rapid and accurate acquisition of soil texture information, this study proposes RVFM, a hybrid deep learning model designed for soil texture detection using microscopic images. The model integrates a CNN branch for extracting multi-dimensional texture features with a Transformer branch for capturing global positional information, fused via a cross-attention module. This architecture effectively captures microscopic distribution characteristics to estimate soil composition proportions. Experimental results demonstrate high precision, with prediction coefficients (R²) for sand, silt, and clay reaching 0.971, 0.954, and 0.931, respectively. Corresponding Root Mean Square Errors (RMSE) were recorded at 3.789, 2.842, and 2.780. The test results outperform those of other classical network models, and the model shows better fitting performance in generalisation tests, demonstrating certain practical value