Abstract:Artificial evaluation of plastic film residue is high labor intensity and low efficiency. A method of combining with color features extraction, impulse coupled neural network segmentation and image morphology algorithm to recognize residual plastic film was proposed in the field by using UAV images. The research area was Pingba County of Anshun City, Guizhou Province, and 1500 images were taken in the research area as experimental data. The UAV was flying at a height of about 40m, and the image data were collected under clear and windfree conditions. These UAV images were conducted geometric correction, 3×3 median filter and histogram equalization processing. Two color space transformation models (RGB, HSV) were compared and analyzed. In order to find out the influence of light intensity on the recognition accuracy, the direct sunlight area and the shadow area of foreground (residual plastic film) and background (soil) were separated to analyze their gray value difference with two color model. It was found that the gray value of shadow area foreground was between the direct sunlight area background and the shadow area background in term of B component while the direct sunlight foreground and shadow area foreground was lower than the background in the term of S component. The manual threshold method, the iterative threshold method, the maximum interclass variance method, the maximum entropy method, the Kmeans clustering method and the impulse coupled neural network were used to segment the residual plastic film from background for both of the B and S components respectively. It was found that the B component was able to recognize sunlight area foreground but not able to recognize shadow area foreground from background. The S component was able to recognize direct sunlight and shadow area foreground from the background. Moreover, the impulse coupled neural network method based on S component had better segmentation effect, and the maximum interclass variance and the iterative threshold method was the second. According to the sunlight direction and different crop growth periods, recognition algorithms for identifying residual film in the field were established. The identification rates were 96.99%, 69.47%, 93.55% and 88.95%, respectively, at sixleaf stage of tobacco growth, after tobacco leaves were harvested, after tobacco rods were pulled out and during the winter idle period. The average overall recognition accuracy of the test area was 87.49%. This method demonstrated fast speed and high recognition accuracy, which can provide a reference for the evaluation and precision collection of residual film.
