Contrast Loss and Detection Performance of Perceptive Sensors in a Controlled Agricultural Dust Chamber

• Custom-built dust chamber developed to study optical perception in dusty air. • Contrast behavior modeled as a function of optical depth. • Object detection accuracy assessed in dusty scenes with a critical threshold. • Dust deposition on the camera window investigated as an additional degradation source. • Beer-Lambert validation shows spatial homogeneity with discrepancies below ten percent. • Airlight effects observed, showing scattered light contribution to contrast loss. Smart farming increasingly relies on optical sensors integrated into agricultural robots and on vision models such as YOLO for perception and navigation. In field conditions, suspended dust scatters and absorbs light, degrading perception data and consequently reducing detection accuracy and operational safety. This study quantifies the impact of agricultural dust on visible-range camera contrast and YOLO-based detection reliability. Experiments were conducted in a custom-built chamber designed to maintain a pseudo-homogeneous dust distribution, ensuring repeatable and reproducible measurements and enabling validation of light attenuation models ranging from Beer-Lambert (BL) to Multiple Scattering (MS) models. The results confirm the applicability of the BL model for characterizing laser light attenuation within optical depths ranging from 0 to 1.5, as the model was evaluated using four transmissometers: three positioned at equal separations and one at a longer separation. Relative discrepancies in the extinction coefficient β were mostly below 10%, indicating limited heterogeneity within the dust chamber. However, a rectangular light source used for contrast measurements induces MS effects, rendering the BL model unsuitable for accurately recovering an unbiased β in this configuration. A strong correlation was observed between the image contrast C ( τ ), and the reference optical depth τ , with C ( τ ) clearly decreasing as τ increased, primarily due to the airlight effect. Fitting a contrast attenuation model accounting for MS airlight confirmed this relationship and revealed an airlight intensity lower than the mean intrinsic intensity of the target, quantifying the contribution of scattered light to contrast degradation. In addition to volumetric attenuation, further contrast loss due to dust deposition on the camera window was quantified. The degradation of YOLO detection performance as a function of τ allows the identification of a threshold level beyond which sensor data becomes unreliable and providing a basis for triggering alerts and adapting system behavior. These results provide a foundation for enhancing the robustness of optical sensors and for developing real-time models capable of predicting visibility loss in agricultural environments using experimentally validated numerical models.