Development of an online monitoring system for cleaning loss in wheat combine harvesters based on CWT-SE-CNN

This study aimed to develop a stable online monitoring system for cleaning loss in wheat combine harvesters. Grain impact signals were also identified under strong machine vibration and mixed separate materials. A double-layer differential piezoelectric sensing and deep learning model were combined to suppress vibration interference using continuous wavelet transform and squeeze-and-excitation attention. Time-frequency features were enhanced for accurate field monitoring of wheat cleaning loss. A double-layer piezoelectric sensor was designed with upper and lower stainless-steel sensing plates equipped with lead zirconate titanate piezoceramic elements. The upper plate was used to collect mixed impact and vibration signals, whereas the lower plate was used to collect background vibration. Two-channel signals were subtracted in the host computer, and butyl-rubber damping components were used to reduce vibration transmission. Impact signals from wheat grains, straw, incompletely threshed wheat ears, and background noise were collected on a bench test rig. The signals were resampled, transformed into time-frequency images using continuous wavelet transformation, and classified using a squeeze-and-excitation convolutional neural network. Huaimai 40 wheat variety was used for dataset construction. A total of 2,400 valid samples were obtained, including 1,200 grain impact samples and 1,200 negative samples composed of straw impact, wheat-ear impact and background-noise samples. The dataset was divided into training, validation and test sets at a ratio of 6:2:2. Each signal segment was normalized to 2,000 data points and converted into a 64 × 64-pixel color time-frequency image using a Morlet wavelet. Three multi-scale convolutional blocks were selected with convolution kernels of 3 × 3, 5 × 5 and 7 × 7. Squeeze-and-excitation modules were introduced to recalibrate channel features and enhance key frequency-band responses. The results showed that the best performance was achieved with an accuracy of 94.2%, a precision of 95.7%, a recall of 92.5% and an F1-score of 94.1% on the test set. The confusion matrix showed that 222 of 240 grain samples and 230 of 240 non-grain samples were identified correctly. Ablation tests showed that removing continuous wavelet transform reduced the accuracy by 2.1 percentage points, removing the squeeze-and-excitation module reduced it by 3.0 percentage points, and replacing continuous wavelet transform with short-time Fourier transform reduced it by 6.9 percentage points. The best overall performance was obtained compared with support vector machine, random forest, multilayer perception, and variable-scale one-dimensional convolutional neural network models. About 1.15 million parameters contained in the model, where the additional parameters introduced by the attention module accounted for less than 0.4% of the total. The average end-to-end processing time was 78.6 ms per frame from raw signal interception to model inference, which was shorter than the 204.8-ms classification and interface-refresh cycle of the host computer. Bench tests further showed that the optimal combination of sensor installation parameters was an inclination angle of 45° and a mounting height of 20 cm above the tail sieve. Detection accuracy decreased slightly as grain flow rate increased, due to the continuous impacts and signal overlap at higher flow rates. Field validation was conducted in Xianggong Town, Linyi City, Shandong Province, using a Kubota 4LZ-4J combine harvester. Nine repeated runs were conducted in the morning, at noon and in the afternoon on the same day to cover crop moisture. The average error rate between system outputs and manually measured cleaning-loss data was 6.57%. The double-layer differential sensor reduced vibration interference, whereas the continuous wavelet transform with squeeze-and-excitation attention improved the extraction and recognition of wheat grain impact features. The bench and field tests verified the high classification accuracy, acceptable computational complexity and near-real-time response. The monitoring system can provide accurate, stable and practical support for online detection of cleaning loss in wheat combined harvesters. The findings can also serve as a technical basis for quality evaluation of intelligent harvesting.